report.object
AI models for financial services
Expert-reviewed from public records, platform data, primary sources, and private research beyond AI search. Use /api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/sample before purchase, then download PDF, JSON, or Markdown after checkout.
page.markdown
AI models for financial services.md
# AI models for financial services
Financial AI models should be compared by accepted output: source authority, model-risk evidence, deployment controls, benchmark caveats, workflow permission and full cost.
## Object
| field | value |
| ---------------- | ------------------------------ |
| report_id | RPT_01KSHN11BR83NYSJ2C5DTNA8E7 |
| schema_version | 2026-05-28 |
| last_updated_at | 2026-06-19 |
| category | MODELS |
| sector | Financial services |
| region | GLOBAL |
| confidence_score | 0.88 |
| source_count | 204 |
| evidence_count | 232 |
## Endpoints
| action | endpoint |
| ------------ | --------------------------------------------------------- |
| summary | /api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7 |
| sample | /api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/sample |
| schema | /api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/schema |
| quote | /api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/quote |
| purchase | /checkout/RPT_01KSHN11BR83NYSJ2C5DTNA8E7?view=agent |
| versions | /api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/versions |
| changelog | /api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/changelog |
| monitor | /api/v1/monitors?report_id=RPT_01KSHN11BR83NYSJ2C5DTNA8E7 |
| mcp_resource | colophon://reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/sample |
## Coverage
| field | value |
| -------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| category | MODELS |
| sector | Financial services |
| region | GLOBAL |
| topics | financial services AI, frontier models, financial RAG, model risk management, AI governance, credit underwriting, investment research, AML and fraud, insurance AI, agentic finance, document intelligence, AI regulation, DORA, EU AI Act, financial benchmarks |
| source_classes | official_model_docs, privacy_terms, case_study, vendor_docs, cloud_docs, model_card, document_ai_docs, financial_benchmark, agent_benchmark, regulator, regulation, risk_framework, policy_framework, financial_model_paper, github, financial_dataset, official_data_source, standard, press_release, research_paper, financial_data_provider, evaluation_framework, status_page, security_advisory, model_hub, benchmark, security_framework, expert_profile, regulatory_sandbox, model_benchmark, market_analysis, regulatory_timeline, regulatory_consultation, pricing_page, market_survey, professional_benchmark, dataset, vendor_research |
| update_cadence | Weekly |
## Sample JSON
```json
{
"report_id": "RPT_01KSHN11BR83NYSJ2C5DTNA8E7",
"slug": "ai-models-financial-services",
"schema_version": "2026-05-28",
"title": "AI models for financial services",
"abstract": "Financial AI models should be compared by accepted output: source authority, model-risk evidence, deployment controls, benchmark caveats, workflow permission and full cost.",
"category": "MODELS",
"sector": "Financial services",
"region": "GLOBAL",
"topics": [
"financial services AI",
"frontier models",
"financial RAG",
"model risk management",
"AI governance",
"credit underwriting",
"investment research",
"AML and fraud",
"insurance AI",
"agentic finance",
"document intelligence",
"AI regulation",
"DORA",
"EU AI Act",
"financial benchmarks"
],
"last_updated_at": "2026-06-19T00:00:00Z",
"source_count": 204,
"evidence_count": 232,
"confidence_score": 0.88,
"web_url": "/r/ai-models-financial-services/RPT_01KSHN11BR83NYSJ2C5DTNA8E7",
"sample_endpoint": "/api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/sample",
"inventory_endpoint": "/api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/inventory",
"full_endpoint": "/api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/full",
"schema_endpoint": "/api/v1/reports/RPT_01KSHN11BR83NYSJ2C5DTNA8E7/schema",
"redaction": {
"status": "sample",
"paid_sections_redacted": true,
"preview_limits": [
"summary",
"table_of_contents",
"evidence_profile"
]
},
"coverage": {
"category": "MODELS",
"sector": "Financial services",
"region": "GLOBAL",
"topics": [
"financial services AI",
"frontier models",
"financial RAG",
"model risk management",
"AI governance",
"credit underwriting",
"investment research",
"AML and fraud",
"insurance AI",
"agentic finance",
"document intelligence",
"AI regulation",
"DORA",
"EU AI Act",
"financial benchmarks"
],
"source_classes": [
"official_model_docs",
"privacy_terms",
"case_study",
"vendor_docs",
"cloud_docs",
"model_card",
"document_ai_docs",
"financial_benchmark",
"agent_benchmark",
"regulator",
"regulation",
"risk_framework",
"policy_framework",
"financial_model_paper",
"github",
"financial_dataset",
"official_data_source",
"standard",
"press_release",
"research_paper",
"financial_data_provider",
"evaluation_framework",
"status_page",
"security_advisory",
"model_hub",
"benchmark",
"security_framework",
"expert_profile",
"regulatory_sandbox",
"model_benchmark",
"market_analysis",
"regulatory_timeline",
"regulatory_consultation",
"pricing_page",
"market_survey",
"professional_benchmark",
"dataset",
"vendor_research"
]
},
"confidence": {
"overall_confidence": 0.88,
"source_quality": 0.9,
"source_recency": 0.92,
"source_density": 0.88,
"claim_verifiability": 0.87,
"geographic_coverage": 0.86,
"known_gaps": [
"Public benchmarks do not fully capture private financial data permissions, document-input condition, customer outcomes or regulator-ready audit exports.",
"Published vendor case studies rarely include denominator data, error rates, reviewer minutes or false-negative rates.",
"Generative and agentic AI treatment in U.S. banking model-risk guidance remains in transition after the 2026 revised guidance.",
"EU AI Act high-risk implementation evidence for financial institutions remains early.",
"AML, sanctions, fraud and market-surveillance AI evidence is sensitive and rarely comparable across institutions.",
"Model-provider terms, retention, price, region support, model versions and tool APIs can change faster than financial-services approval cycles.",
"Third-party concentration risk from common cloud, model and data-provider routes is hard to quantify from public evidence.",
"Open-weight model economics depend on private infrastructure utilization, security controls and validation cost.",
"Finance-specific models such as BloombergGPT, FinGPT and FinBERT prove domain-data value but do not provide current licensed data or route approval by themselves.",
"Source entitlements, identifier mapping, parser quality and data currentness are rarely visible in public benchmark results.",
"Case studies show workflow shape but usually omit false negatives, severity-weighted errors, regulator review and control-group results.",
"Operational signals such as status pages, CVEs and model-hub changes remain indirect but important route-reapproval triggers."
]
},
"document": {
"object": "report_document",
"schemaVersion": "2026-05-28",
"title": "AI models for financial services",
"subtitle": "Approved financial outputs beat clever models.",
"abstract": "Financial AI models should be compared by accepted output: source authority, model-risk evidence, deployment controls, benchmark caveats, workflow permission and full cost.",
"category": "MODELS",
"sector": "Financial services",
"region": "GLOBAL",
"topics": [
"financial services AI",
"frontier models",
"financial RAG",
"model risk management",
"AI governance",
"credit underwriting",
"investment research",
"AML and fraud",
"insurance AI",
"agentic finance",
"document intelligence",
"AI regulation",
"DORA",
"EU AI Act",
"financial benchmarks"
],
"sources": [
{
"id": "S1",
"label": "OpenAI API model docs",
"sourceClass": "official_model_docs",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://platform.openai.com/docs/models"
},
{
"id": "S2",
"label": "OpenAI enterprise privacy",
"sourceClass": "privacy_terms",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://openai.com/enterprise-privacy"
},
{
"id": "S3",
"label": "OpenAI and Morgan Stanley case study",
"sourceClass": "case_study",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://openai.com/index/morgan-stanley/"
},
{
"id": "S4",
"label": "Anthropic model overview",
"sourceClass": "official_model_docs",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://docs.anthropic.com/en/docs/about-claude/models/overview"
},
{
"id": "S5",
"label": "Anthropic - agents for financial services",
"sourceClass": "vendor_docs",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://www.anthropic.com/news/finance-agents"
},
{
"id": "S6",
"label": "Anthropic commercial privacy terms",
"sourceClass": "privacy_terms",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://support.anthropic.com/en/articles/9267385-does-anthropic-act-as-a-data-processor-or-controller"
},
{
"id": "S7",
"label": "Google Cloud Vertex AI documentation",
"sourceClass": "cloud_docs",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://docs.cloud.google.com/vertex-ai/docs"
},
{
"id": "S8",
"label": "Google Cloud Vertex AI data governance",
"sourceClass": "privacy_terms",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://cloud.google.com/vertex-ai/generative-ai/docs/data-governance"
},
{
"id": "S9",
"label": "Amazon Bedrock overview",
"sourceClass": "cloud_docs",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://aws.amazon.com/bedrock/"
},
{
"id": "S10",
"label": "Microsoft Azure OpenAI data, privacy, and security",
"sourceClass": "privacy_terms",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://learn.microsoft.com/en-us/legal/cognitive-services/openai/data-privacy"
},
{
"id": "S11",
"label": "Meta Llama model access",
"sourceClass": "model_card",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://ai.meta.com/llama/get-started/"
},
{
"id": "S12",
"label": "Mistral model overview",
"sourceClass": "official_model_docs",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://docs.mistral.ai/models/overview"
},
{
"id": "S13",
"label": "Mistral Document AI OCR",
"sourceClass": "document_ai_docs",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://docs.mistral.ai/capabilities/document_ai/basic_ocr/"
},
{
"id": "S14",
"label": "Qwen3 embedding and reranker",
"sourceClass": "model_card",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://qwenlm.github.io/blog/qwen3-embedding/"
},
{
"id": "S15",
"label": "Cohere Rerank documentation",
"sourceClass": "vendor_docs",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://docs.cohere.com/v2/docs/rerank"
},
{
"id": "S16",
"label": "Voyage AI voyage-3.5 release",
"sourceClass": "model_card",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "sample",
"transformChain": [],
"url": "https://blog.voyageai.com/2025/06/03/voyage-3-5/"
},
{
"id": "S20",
"label": "FinanceBench paper",
"sourceClass": "financial_benchmark",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "locked",
"transformChain": [],
"url": "https://arxiv.org/abs/2311.11944"
},
{
"id": "S26",
"label": "OCC Bulletin 2026-13",
"sourceClass": "regulator",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "locked",
"transformChain": [],
"url": "https://occ.gov/news-issuances/bulletins/2026/bulletin-2026-13.html"
},
{
"id": "S29",
"label": "Bank of England - AI in the financial system",
"sourceClass": "regulator",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "locked",
"transformChain": [],
"url": "https://www.bankofengland.co.uk/financial-stability-in-focus/2025/april-2025"
},
{
"id": "S31",
"label": "FINRA Regulatory Notice 24-09",
"sourceClass": "regulator",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "locked",
"transformChain": [],
"url": "https://www.finra.org/rules-guidance/notices/24-09"
},
{
"id": "S37",
"label": "EU AI Act Service Desk - Annex III",
"sourceClass": "regulation",
"accessedAt": "2026-06-19",
"status": "verified",
"access": "locked",
"transformChain": [],
"url": "https://ai-act-service-desk.ec.europa.eu/en/ai-act/annex-3"
}
],
"evidence": [
{
"id": "E1",
"sourceId": "S1",
"claimIds": [
"C1",
"C2"
],
"summary": "OpenAI's model catalogue positions model choice as a product and API routing decision across reasoning effort, cost, latency, multimodal input and tool use.",
"status": "verified",
"confidence": 0.88,
"access": "sample"
},
{
"id": "E3",
"sourceId": "S3",
"claimIds": [
"C3"
],
"summary": "OpenAI's Morgan Stanley case study emphasizes private wealth-management content, evaluation practice and domain workflow integration rather than a generic chatbot rollout.",
"status": "verified",
"confidence": 0.88,
"access": "sample"
},
{
"id": "E20",
"sourceId": "S20",
"claimIds": [
"C3",
"C7"
],
"summary": "FinanceBench reports 10,231 public-company financial QA items with answers and evidence strings, built as an open-book financial QA benchmark.",
"status": "verified",
"confidence": 0.88,
"access": "sample"
},
{
"id": "E21",
"sourceId": "S20",
"claimIds": [
"C3",
"C7"
],
"summary": "FinanceBench found GPT-4-Turbo with retrieval incorrectly answered or refused 81% of sampled questions, making financial RAG quality a production issue.",
"status": "verified",
"confidence": 0.88,
"access": "sample"
},
{
"id": "E28",
"sourceId": "S26",
"claimIds": [
"C6"
],
"summary": "OCC Bulletin 2026-13 replaced older model-risk guidance for covered banking organizations and expressly leaves generative and agentic AI outside that guidance's scope.",
"status": "verified",
"confidence": 0.88,
"access": "sample"
},
{
"id": "E32",
"sourceId": "S29",
"claimIds": [],
"summary": "The Bank of England's Financial Stability in Focus paper treats AI adoption as a potential financial-stability issue when common models, data or vendors influence core financial decisions.",
"status": "verified",
"confidence": 0.88,
"access": "sample"
},
{
"id": "E35",
"sourceId": "S31",
"claimIds": [
"C6"
],
"summary": "FINRA Regulatory Notice 24-09 reminds member firms that securities laws and FINRA rules remain technology neutral when firms use generative AI or LLMs.",
"status": "verified",
"confidence": 0.88,
"access": "sample"
},
{
"id": "E42",
"sourceId": "S37",
"claimIds": [
"C6"
],
"summary": "EU AI Act Annex III lists creditworthiness and credit scoring for natural persons, excluding fraud detection, as high-risk AI systems.",
"status": "verified",
"confidence": 0.88,
"access": "sample"
}
],
"claims": [
{
"id": "C1",
"statement": "Financial AI model selection is an approval decision: task, data class, source path, model lane, hosting route, reviewer and approval gate.",
"confidence": 0.91,
"evidenceIds": [
"E1",
"E3"
],
"access": "sample"
},
{
"id": "C2",
"statement": "Frontier models belong on the shortlist for research synthesis, drafting, reasoning and workflow orchestration, but finance buyers should route them by task and control plane.",
"confidence": 0.9,
"evidenceIds": [
"E1"
],
"access": "sample"
},
{
"id": "C3",
"statement": "The hard part of financial AI is grounded numerical work over filings, tables, time periods and private records, not fluent prose.",
"confidence": 0.92,
"evidenceIds": [
"E20",
"E21"
],
"access": "sample"
},
{
"id": "C6",
"statement": "Regulated finance turns model choice into evidence: data rights, explainability, human oversight, third-party risk, audit logs, change control and conduct compliance.",
"confidence": 0.93,
"evidenceIds": [
"E28",
"E35",
"E42"
],
"access": "sample"
},
{
"id": "C7",
"statement": "Public finance and professional-document benchmarks expose failure modes; private approval still needs the buyer's own sources, permissions, reviewer rules and audit artefacts.",
"confidence": 0.9,
"evidenceIds": [
"E20",
"E21"
],
"access": "sample"
}
],
"colophon": {
"checkedAt": "2026-06-19",
"latestRevisionAt": "2026-06-19T00:00:00Z",
"reviewStandard": "expert_human_reviewed",
"kinetics": {
"speed": "fast",
"impact": 0.89,
"momentum": 0.86,
"businessImpact": "Finance model approval has shifted from choosing an LLM to proving the evidence route can survive review.",
"rateOfChange": "AI Act consultations, GPAI enforcement and finance benchmark refreshes can change credit, insurance, filing-RAG and agent routes before renewal.",
"from": "general chat pilots",
"toward": "routed finance stacks with source controls, document intelligence, model-risk evidence, tool gates and human review"
},
"watch": [
"23 June 2026: European Commission high-risk AI consultation closes; credit and insurance route classification may move [E180].",
"2 August 2026: Commission GPAI enforcement powers begin; model documentation, transparency and provider-diligence evidence may change [E181]."
],
"gapAnalysis": "Comparable public benchmarks, downstream parser-to-answer tests, longitudinal product telemetry, permissioned-corpus datasets, AML and fraud false-negative datasets, customer-outcome studies, model-change regression logs, concentration-event data and audit-export evidence remain missing for financial model routes."
},
"body": {
"sections": [
{
"id": "verdict",
"title": "Executive summary",
"access": "sample",
"claimIds": [
"C1",
"C2",
"C3",
"C6",
"C7"
],
"evidenceIds": [
"E1",
"E3",
"E20",
"E21",
"E28",
"E32",
"E35",
"E42"
],
"blocks": [
{
"id": "verdict-narrative",
"evidenceIds": [
"E1",
"E3",
"E20",
"E21",
"E28",
"E32",
"E35",
"E42"
],
"type": "section_narrative",
"lead": "The best financial AI system is not the cleverest model. It is the control stack that produces a reviewable answer from approved sources under the right control regime. A strong frontier model with a stale filing, a broken table or an unapproved customer-data path is still a bad financial answer.",
"body": [
"Use frontier models where judgement, drafting, synthesis and multi-step reasoning matter. Use cheaper, private or open-weight models where throughput, sovereignty, latency or custom controls dominate. Use OCR, document intelligence, embeddings and rerankers before asking any model to answer a question over filings, KYC packs, statements, credit files or policy documents.",
"FinanceBench gives the warning in numbers: its original retrieval baseline incorrectly answered or refused 81% of sampled financial QA questions [E21]. That result is old by model standards, but the lesson has aged well. Retrieval, tables, dates and evidence chains can break a finance workflow before generation begins."
],
"detail": [
"Rank model lanes before ranking model names. Investment research, retail credit, AML investigation, insurance pricing, treasury analysis, month-end close and customer communications need different routes.",
"The procurement object is therefore the approved route: data class, source path, parser, retriever, generator, hosting route, reviewer, logs, contract terms and reapproval trigger."
]
},
{
"id": "banking-markets-insurance-report-link",
"evidenceIds": [],
"type": "report_link",
"reportId": "RPT_01KSHN11BSWF9E8EAJ1WVS9FG4",
"slug": "ai-models-banking-capital-markets-insurance",
"label": "AI models for banking, capital markets, and insurance",
"dek": "Use the regulated-finance report for banking, capital-markets and insurance route approval, customer-impact gates, securities supervision, insurance controls and demand-signal split timing.",
"relation": "Related report"
},
{
"id": "financial-file-processing-report-link",
"evidenceIds": [],
"type": "report_link",
"reportId": "RPT_01KSHN11BQRTZBGKKP66HHC0FX",
"slug": "file-processing-providers-financial-ai",
"label": "File processing providers for financial AI",
"dek": "Detailed due diligence for financial OCR, document AI, parser routes, bank statements, filings, invoices, IDs, claims files, controls, pricing and parser evals.",
"relation": "Related report"
},
{
"id": "executive-table",
"title": "Board-level verdict",
"evidenceIds": [
"E1",
"E3",
"E20",
"E21",
"E28",
"E32",
"E35",
"E42"
],
"type": "table",
"columns": [
{
"key": "question",
"label": "Question",
"align": "left"
},
{
"key": "answer",
"label": "Answer",
"align": "left"
},
{
"key": "buyer_move",
"label": "Buyer move",
"align": "left"
}
],
"rows": [
{
"question": "Which model wins?",
"answer": "The winning route changes by task, data class and approval gate.",
"buyer_move": "Approve model lanes, not model names."
},
{
"question": "Where is the main failure?",
"answer": "Wrong source, lost table, stale date, weak calculation or unsupported claim.",
"buyer_move": "Test source capture, retrieval, calculation and citation before model prose."
},
{
"question": "Where should frontier models lead?",
"answer": "Research synthesis, drafting, reasoning, workflow orchestration and exception analysis.",
"buyer_move": "Pair them with verified sources, deterministic calculations and reviewer gates."
},
{
"question": "Where should cheaper or private models lead?",
"answer": "Classification, extraction, volume summarization, internal search and low-risk finance operations.",
"buyer_move": "Use them where measured quality clears the workflow and data controls are easier."
}
]
},
{
"id": "route-definition-callout",
"title": "Route definition",
"evidenceIds": [
"E1",
"E3",
"E20",
"E21",
"E28",
"E32",
"E35",
"E42"
],
"type": "callout",
"tone": "neutral",
"body": "Route means the approved path from work to accepted output: the source set, data boundary, model or vendor channel, tools, reviewer, evidence record and reapproval trigger that make an AI result usable."
},
{
"id": "verdict-section-callout",
"title": "Source before you rank",
"evidenceIds": [
"E1",
"E3",
"E20",
"E21",
"E28",
"E32",
"E35",
"E42"
],
"type": "callout",
"tone": "info",
"body": "Approve the financial AI route before the model: source rights, parser, retriever, generator, tool permissions, reviewer, logs and reapproval trigger."
}
]
},
{
"id": "decision-map",
"title": "Decision map",
"access": "sample",
"dek": "Approve outputs by work type and risk class.",
"claimIds": [
"C1",
"C6"
],
"evidenceIds": [
"E1",
"E3",
"E35",
"E42"
],
"blocks": [
{
"id": "decision-map-narrative",
"evidenceIds": [
"E1",
"E35",
"E42"
],
"type": "section_narrative",
"lead": "Financial AI buying fails when one approval covers unlike work.",
"body": [
"The first split is not model brand. It is whether the system informs an employee, drafts a customer-facing artefact, influences a regulated decision or acts through a tool.",
"A filing-summary assistant can tolerate different controls from a credit-scoring model. An AML investigator's summarizer is not the same as an agent that files a suspicious-activity draft. A portfolio-research copilot is not the same as an execution system."
],
"detail": [
"Build the route register around intended use. Each route should name the work, source corpus, data class, approved model lane, hosting route, tool permissions, reviewer, logging standard, evidence export and reapproval trigger.",
"The route register becomes the bridge between model-risk, technology-risk, compliance, legal, security and business owners."
]
},
{
"id": "route-decision-tree",
"evidenceIds": [
"E35",
"E42"
],
"type": "decision_tree",
"lead": "Start with the consequence of the output.",
"steps": [
{
"prompt": "Does the output affect credit, insurance pricing, advice, trading, surveillance or customer treatment?",
"tone": "warning",
"branches": [
{
"label": "Yes",
"outcome": "Regulated decision route.",
"next": "Run legal, compliance, model-risk and fairness gates.",
"tone": "blocked"
},
{
"label": "No",
"outcome": "Continue.",
"next": "Check data class and tool permissions.",
"tone": "neutral"
}
]
},
{
"prompt": "Does the route use non-public customer, account, trading, portfolio or employee data?",
"tone": "warning",
"branches": [
{
"label": "Yes",
"outcome": "Controlled data path.",
"next": "Approve retention, region, training use, access logs and subprocessors.",
"tone": "warning"
},
{
"label": "No",
"outcome": "Continue.",
"next": "Check whether the system can act.",
"tone": "neutral"
}
]
},
{
"prompt": "Can the system write, submit, send, approve, trade, update records or mutate spreadsheets?",
"tone": "warning",
"branches": [
{
"label": "Yes",
"outcome": "Agent route.",
"next": "Use tool permissions, dry-run mode, action logs and human approval.",
"tone": "warning"
},
{
"label": "No",
"outcome": "Assistant route.",
"next": "Run task evals and review logging before scale.",
"tone": "success"
}
]
}
]
},
{
"id": "route-register-table",
"title": "Minimum approval register",
"evidenceIds": [
"E1",
"E35",
"E42"
],
"type": "table",
"columns": [
{
"key": "field",
"label": "Field",
"align": "left"
},
{
"key": "why_it_matters",
"label": "Why it matters",
"align": "left"
},
{
"key": "bad_entry",
"label": "Bad entry",
"align": "left"
},
{
"key": "good_entry",
"label": "Good entry",
"align": "left"
}
],
"rows": [
{
"field": "Use case",
"why_it_matters": "Regulation follows intended use.",
"bad_entry": "Finance AI.",
"good_entry": "Retail-credit adverse-action explanation draft."
},
{
"field": "Source path",
"why_it_matters": "The model cannot reason over missing evidence.",
"bad_entry": "Documents.",
"good_entry": "SEC filings, policy KB, credit file, account events and reviewer notes."
},
{
"field": "Data class",
"why_it_matters": "Retention, region and access controls change.",
"bad_entry": "Internal data.",
"good_entry": "Public filings plus confidential customer account data."
},
{
"field": "Model lane",
"why_it_matters": "Capability, cost and approval differ.",
"bad_entry": "GPT.",
"good_entry": "Frontier API for reasoning; private classifier for triage; reranker for retrieval."
},
{
"field": "Reviewer",
"why_it_matters": "Accountability needs a named human role.",
"bad_entry": "Human review.",
"good_entry": "Credit analyst signs off reason codes before customer notice."
},
{
"field": "Reapproval trigger",
"why_it_matters": "Model drift is a governance event.",
"bad_entry": "Annual review.",
"good_entry": "Model version, prompt, data source, threshold, tool permission or vendor term change."
}
]
},
{
"id": "decision-map-section-callout",
"title": "One bank, many control paths",
"evidenceIds": [
"E1",
"E3",
"E35",
"E42"
],
"type": "callout",
"tone": "info",
"body": "A model that is acceptable for analyst drafting may be blocked for credit scoring, conditional for KYC, and useful for internal policy search."
}
]
},
{
"id": "what-changed",
"title": "What changed",
"access": "sample",
"dek": "Financial AI moved from productivity demos to output governance.",
"claimIds": [
"C6"
],
"evidenceIds": [],
"blocks": [
{
"id": "what-changed-narrative",
"evidenceIds": [],
"type": "section_narrative",
"lead": "The market has stopped asking whether AI can summarize finance work. It is asking whether AI can touch regulated decisions, customer records, private data, market infrastructure and financial crime workflows without losing evidence.",
"body": [
"Three changes forced the shift. First, model capability is now strong enough to enter real workflows. Second, domain and source access matter more because financial work depends on current filings, prices, entity identities, sanctions records, policy documents and private ledgers. Third, regulators are turning AI into an inventory, governance, lifecycle and resilience question.",
"BloombergGPT and FinGPT made the same point from different directions: finance-specific data matters. BloombergGPT used privileged financial corpora; FinGPT argued for open financial-data pipelines and adaptation. Neither route removes the need for current sources, permissions and private evals [E73, E75]."
],
"detail": [
"The institutional consequence is a new approval object: the model route. A bank can approve a model for analyst research and block it for credit decisions. An insurer can approve document extraction for claims support and block autonomous pricing. An asset manager can approve filing RAG and still prohibit unreviewed client communications.",
"The old pilot question was, 'Does the answer look good?' The production question is, 'Can we prove where it came from, who approved it, what it changed, and why it should still be trusted after the next model update?'"
]
},
{
"id": "what-changed-table",
"title": "From pilot to production",
"evidenceIds": [],
"type": "table",
"columns": [
{
"key": "old_question",
"label": "Old question",
"align": "left"
},
{
"key": "new_question",
"label": "New question",
"align": "left"
},
{
"key": "finance_implication",
"label": "Finance implication",
"align": "left"
}
],
"rows": [
{
"old_question": "Which model is smartest?",
"new_question": "Which approved route clears the task?",
"finance_implication": "Model, data, tools and reviewer move together."
},
{
"old_question": "Can it summarize a filing?",
"new_question": "Can it cite the right filing, table, period and calculation?",
"finance_implication": "Source integrity beats prose quality."
},
{
"old_question": "Can employees use chat?",
"new_question": "Which records, customers and jurisdictions enter the prompt?",
"finance_implication": "Data class decides deployment."
},
{
"old_question": "Can the agent save time?",
"new_question": "Can the agent act, and who approves the action?",
"finance_implication": "Tool permissions become controls."
},
{
"old_question": "Is the vendor secure?",
"new_question": "Can the institution exit, audit and survive concentration?",
"finance_implication": "Third-party resilience is part of model selection."
}
]
},
{
"id": "what-changed-section-callout",
"title": "The pilot era is over",
"evidenceIds": [],
"type": "callout",
"tone": "info",
"body": "The live question is no longer whether finance will use AI. It is which routes may touch which data, decisions, customers and systems."
}
]
},
{
"id": "model-economics",
"title": "Model economics",
"access": "sample",
"dek": "Price the finished financial task, not the token.",
"claimIds": [
"C3",
"C7"
],
"evidenceIds": [
"E20",
"E21"
],
"blocks": [
{
"id": "model-economics-narrative",
"evidenceIds": [
"E20",
"E21"
],
"type": "section_narrative",
"lead": "The right economic unit is not a token. It is a signed-off answer, cell, memo, alert or customer communication.",
"body": [
"Financial work has a verification tail. A model answer often needs source inspection, recalculation, variance checks, reviewer notes, policy fit, approval and record retention.",
"That tail can dominate model price. A cheaper model that sends reviewers to the filing five extra times loses to a more expensive model that returns the right table, period and calculation with a usable citation.",
"Artificial Analysis-style comparisons are useful because they put price beside intelligence, output speed, latency and context [E177, E178]. Financial buyers need one more column: accepted output. The unit is not a million tokens. It is a reviewed answer, reconciled cell, signed-off memo, closed alert or archived customer communication.",
"Silicon Valley Bank's State of Enterprise Software survey shows why that denominator matters. About 48% of respondents reported 25%+ employee-productivity improvement from AI, but half said revenue-growth attribution was hard to measure and one-quarter put the revenue lift below 10% [E226]. Treat productivity and revenue as separate economic claims."
],
"detail": [
"FinanceBench and agent benchmarks move evaluation away from general language skill and toward finance-specific failure modes: finding the right evidence, doing the arithmetic, preserving dates, submitting an answer and surviving expert review.",
"Procurement should therefore run a cost-per-accepted-output eval. Count false answers, refusals, missing-source reports, reviewer minutes, reruns, latency, token spend, tool calls and audit-export quality.",
"Pricing is moving for the same reason. Silicon Valley Bank reports that nearly 40% of surveyed enterprise-software companies are primarily seat-priced today, but only 26% plan to stay there; customer demand and AI/compute economics are explicit pricing drivers [E227]. Do not let the vendor's billing unit become the buyer's ROI unit."
]
},
{
"id": "intelligence-price-decline-chart-split",
"title": "Token price misses finance work",
"evidenceIds": [],
"type": "chart",
"chartType": "line",
"data": [
{
"date": "2024-03-04",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.1,
"valueLabel": "$0.1",
"model": "Llama 2 Chat 7B",
"observed": false,
"current": false
},
{
"date": "2024-04-18",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.07,
"valueLabel": "$0.07",
"model": "Llama 3 Instruct 8B",
"observed": true,
"current": false
},
{
"date": "2024-09-25",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.05,
"valueLabel": "$0.05",
"model": "Llama 3.2 Instruct 1B",
"observed": true,
"current": false
},
{
"date": "2025-06-26",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.025,
"valueLabel": "$0.025",
"model": "Gemma 3n E4B Instruct",
"observed": true,
"current": false
},
{
"date": "2026-03-02",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.02,
"valueLabel": "$0.02",
"model": "Qwen3.5 0.8B (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.02,
"valueLabel": "$0.02",
"model": "Qwen3.5 0.8B (Non-reasoning)",
"observed": false,
"current": true
},
{
"date": "2024-03-04",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.15,
"valueLabel": "$0.15",
"model": "Solar Mini",
"observed": false,
"current": false
},
{
"date": "2024-07-23",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.1,
"valueLabel": "$0.1",
"model": "Llama 3.1 Instruct 8B",
"observed": true,
"current": false
},
{
"date": "2024-11-18",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.088,
"valueLabel": "$0.088",
"model": "Qwen2.5 Turbo",
"observed": true,
"current": false
},
{
"date": "2024-12-03",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.061,
"valueLabel": "$0.061",
"model": "Nova Micro",
"observed": true,
"current": false
},
{
"date": "2026-02-25",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.052,
"valueLabel": "$0.052",
"model": "LFM2 24B A2B",
"observed": true,
"current": false
},
{
"date": "2026-03-02",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.02,
"valueLabel": "$0.02",
"model": "Qwen3.5 0.8B (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.02,
"valueLabel": "$0.02",
"model": "Qwen3.5 0.8B (Reasoning)",
"observed": false,
"current": true
},
{
"date": "2024-09-12",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 28.875,
"valueLabel": "$28.9",
"model": "o1-preview",
"observed": true,
"current": false
},
{
"date": "2025-01-31",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 1.925,
"valueLabel": "$1.93",
"model": "o3-mini (high)",
"observed": true,
"current": false
},
{
"date": "2025-03-25",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 1.209,
"valueLabel": "$1.21",
"model": "DeepSeek V3 0324",
"observed": true,
"current": false
},
{
"date": "2025-04-14",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.7,
"valueLabel": "$0.7",
"model": "GPT-4.1 mini",
"observed": true,
"current": false
},
{
"date": "2025-07-21",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.356,
"valueLabel": "$0.356",
"model": "Qwen3 235B A22B 2507 Instruct",
"observed": true,
"current": false
},
{
"date": "2025-07-31",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.352,
"valueLabel": "$0.352",
"model": "Qwen3 Coder 30B A3B Instruct",
"observed": true,
"current": false
},
{
"date": "2025-08-05",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.088,
"valueLabel": "$0.088",
"model": "gpt-oss-20B (high)",
"observed": true,
"current": false
},
{
"date": "2026-03-02",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.06,
"valueLabel": "$0.06",
"model": "Qwen3.5 4B (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.06,
"valueLabel": "$0.06",
"model": "Qwen3.5 4B (Non-reasoning)",
"observed": false,
"current": true
},
{
"date": "2024-12-05",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 26.25,
"valueLabel": "$26.3",
"model": "o1",
"observed": true,
"current": false
},
{
"date": "2025-02-19",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.35,
"valueLabel": "$0.35",
"model": "Grok 3 mini Reasoning (high)",
"observed": true,
"current": false
},
{
"date": "2025-08-05",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.262,
"valueLabel": "$0.262",
"model": "gpt-oss-120b (high)",
"observed": true,
"current": false
},
{
"date": "2025-12-16",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.15,
"valueLabel": "$0.15",
"model": "MiMo-V2-Flash (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-03-02",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.113,
"valueLabel": "$0.113",
"model": "Qwen3.5 9B (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.113,
"valueLabel": "$0.113",
"model": "Qwen3.5 9B (Reasoning)",
"observed": false,
"current": true
},
{
"date": "2025-06-10",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 35,
"valueLabel": "$35",
"model": "o3-pro",
"observed": true,
"current": false
},
{
"date": "2025-07-10",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 11,
"valueLabel": "$11",
"model": "Grok 4",
"observed": true,
"current": false
},
{
"date": "2025-08-07",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 0.688,
"valueLabel": "$0.688",
"model": "GPT-5 mini (high)",
"observed": true,
"current": false
},
{
"date": "2025-12-01",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 0.337,
"valueLabel": "$0.337",
"model": "DeepSeek V3.2 (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2025-12-16",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 0.15,
"valueLabel": "$0.15",
"model": "MiMo-V2-Flash (Feb 2026)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 0.15,
"valueLabel": "$0.15",
"model": "MiMo-V2-Flash (Feb 2026)",
"observed": false,
"current": true
},
{
"date": "2025-12-11",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 4.813,
"valueLabel": "$4.81",
"model": "GPT-5.2 (xhigh)",
"observed": true,
"current": false
},
{
"date": "2026-02-19",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 4.5,
"valueLabel": "$4.5",
"model": "Gemini 3.1 Pro Preview",
"observed": true,
"current": false
},
{
"date": "2026-04-02",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 1.125,
"valueLabel": "$1.13",
"model": "Qwen3.6 Plus",
"observed": true,
"current": false
},
{
"date": "2026-04-24",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 0.544,
"valueLabel": "$0.544",
"model": "DeepSeek V4 Pro (Reasoning, Max Effort)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 0.544,
"valueLabel": "$0.544",
"model": "DeepSeek V4 Pro (Reasoning, Max Effort)",
"observed": false,
"current": true
},
{
"date": "2025-03-12",
"seriesId": "aa-0",
"seriesLabel": "Intelligence <10",
"metric": "aa",
"panel": "bill",
"tierIndex": 0,
"value": 6.518,
"valueLabel": "$6.52",
"model": "Gemma 3 4B Instruct",
"observed": true,
"current": false
},
{
"date": "2025-06-26",
"seriesId": "aa-0",
"seriesLabel": "Intelligence <10",
"metric": "aa",
"panel": "bill",
"tierIndex": 0,
"value": 5.7215,
"valueLabel": "$5.72",
"model": "Gemma 3n E4B Instruct",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-0",
"seriesLabel": "Intelligence <10",
"metric": "aa",
"panel": "bill",
"tierIndex": 0,
"value": 5.7215,
"valueLabel": "$5.72",
"model": "Gemma 3n E4B Instruct",
"observed": false,
"current": true
},
{
"date": "2024-03-04",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 36.4377,
"valueLabel": "$36.4",
"model": "Claude 3 Haiku",
"observed": true,
"current": false
},
{
"date": "2024-07-23",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 27.7078,
"valueLabel": "$27.7",
"model": "Llama 3.1 Instruct 8B",
"observed": true,
"current": false
},
{
"date": "2024-12-03",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 6.3848,
"valueLabel": "$6.38",
"model": "Nova Micro",
"observed": true,
"current": false
},
{
"date": "2025-09-22",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 4.4809,
"valueLabel": "$4.48",
"model": "Granite 4.0 H Small",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 4.4809,
"valueLabel": "$4.48",
"model": "Granite 4.0 H Small",
"observed": false,
"current": true
},
{
"date": "2025-01-31",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 314.7215,
"valueLabel": "$315",
"model": "o3-mini (high)",
"observed": true,
"current": false
},
{
"date": "2025-03-25",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 59.2474,
"valueLabel": "$59.2",
"model": "DeepSeek V3 0324",
"observed": true,
"current": false
},
{
"date": "2025-04-14",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 53.7334,
"valueLabel": "$53.7",
"model": "GPT-4.1 mini",
"observed": true,
"current": false
},
{
"date": "2025-07-31",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 29.2907,
"valueLabel": "$29.3",
"model": "Qwen3 Coder 30B A3B Instruct",
"observed": true,
"current": false
},
{
"date": "2025-08-05",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 7.6768,
"valueLabel": "$7.68",
"model": "gpt-oss-20B (low)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 7.6768,
"valueLabel": "$7.68",
"model": "gpt-oss-20B (low)",
"observed": false,
"current": true
},
{
"date": "2024-12-05",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 3270.3573,
"valueLabel": "$3,270",
"model": "o1",
"observed": true,
"current": false
},
{
"date": "2025-02-19",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 104.8919,
"valueLabel": "$105",
"model": "Grok 3 mini Reasoning (high)",
"observed": true,
"current": false
},
{
"date": "2025-08-05",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 67.3659,
"valueLabel": "$67.4",
"model": "gpt-oss-120b (high)",
"observed": true,
"current": false
},
{
"date": "2025-08-07",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 61.9119,
"valueLabel": "$61.9",
"model": "GPT-5 mini (medium)",
"observed": true,
"current": false
},
{
"date": "2025-09-19",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 35.9855,
"valueLabel": "$36",
"model": "Grok 4 Fast (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2025-12-16",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 21.3815,
"valueLabel": "$21.4",
"model": "MiMo-V2-Flash (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-04-02",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 19.4332,
"valueLabel": "$19.4",
"model": "Gemma 4 31B (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 19.4332,
"valueLabel": "$19.4",
"model": "Gemma 4 31B (Non-reasoning)",
"observed": false,
"current": true
},
{
"date": "2025-07-10",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 2881.1519,
"valueLabel": "$2,881",
"model": "Grok 4",
"observed": true,
"current": false
},
{
"date": "2025-08-07",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 168.4577,
"valueLabel": "$168",
"model": "GPT-5 mini (high)",
"observed": true,
"current": false
},
{
"date": "2025-12-01",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 75.6823,
"valueLabel": "$75.7",
"model": "DeepSeek V3.2 (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2025-12-16",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 66.5448,
"valueLabel": "$66.5",
"model": "MiMo-V2-Flash (Feb 2026)",
"observed": true,
"current": false
},
{
"date": "2026-04-24",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 57.4318,
"valueLabel": "$57.4",
"model": "DeepSeek V4 Flash (Reasoning, High Effort)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 57.4318,
"valueLabel": "$57.4",
"model": "DeepSeek V4 Flash (Reasoning, High Effort)",
"observed": false,
"current": true
},
{
"date": "2025-12-11",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 2304.2197,
"valueLabel": "$2,304",
"model": "GPT-5.2 (xhigh)",
"observed": true,
"current": false
},
{
"date": "2026-02-05",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 1571.5554,
"valueLabel": "$1,572",
"model": "GPT-5.3 Codex (xhigh)",
"observed": true,
"current": false
},
{
"date": "2026-02-19",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 892.2838,
"valueLabel": "$892",
"model": "Gemini 3.1 Pro Preview",
"observed": true,
"current": false
},
{
"date": "2026-04-02",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 482.6505,
"valueLabel": "$483",
"model": "Qwen3.6 Plus",
"observed": true,
"current": false
},
{
"date": "2026-04-22",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 461.5912,
"valueLabel": "$462",
"model": "MiMo-V2.5-Pro",
"observed": true,
"current": false
},
{
"date": "2026-04-24",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 267.8209,
"valueLabel": "$268",
"model": "DeepSeek V4 Pro (Reasoning, Max Effort)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 267.8209,
"valueLabel": "$268",
"model": "DeepSeek V4 Pro (Reasoning, Max Effort)",
"observed": false,
"current": true
}
],
"spec": {
"$schema": "https://vega.github.io/schema/vega-lite/v6.json",
"config": {
"background": null,
"view": {
"stroke": null
},
"axis": {
"domain": false,
"tickColor": "#c8c8c0"
}
},
"vconcat": [
{
"hconcat": [
{
"title": {
"text": "Sticker price per million tokens (blended)",
"anchor": "start",
"font": "IBM Plex Mono",
"fontSize": 18,
"offset": 12,
"color": "#111111"
},
"width": 470,
"height": 320,
"padding": {
"top": 0,
"right": 44,
"bottom": 0,
"left": 0
},
"data": {
"values": [
{
"date": "2024-03-04",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.1,
"valueLabel": "$0.1",
"model": "Llama 2 Chat 7B",
"observed": false,
"current": false
},
{
"date": "2024-04-18",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.07,
"valueLabel": "$0.07",
"model": "Llama 3 Instruct 8B",
"observed": true,
"current": false
},
{
"date": "2024-09-25",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.05,
"valueLabel": "$0.05",
"model": "Llama 3.2 Instruct 1B",
"observed": true,
"current": false
},
{
"date": "2025-06-26",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.025,
"valueLabel": "$0.025",
"model": "Gemma 3n E4B Instruct",
"observed": true,
"current": false
},
{
"date": "2026-03-02",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.02,
"valueLabel": "$0.02",
"model": "Qwen3.5 0.8B (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-0",
"seriesLabel": "Intelligence <10",
"metric": "token",
"panel": "token",
"tierIndex": 0,
"value": 0.02,
"valueLabel": "$0.02",
"model": "Qwen3.5 0.8B (Non-reasoning)",
"observed": false,
"current": true
},
{
"date": "2024-03-04",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.15,
"valueLabel": "$0.15",
"model": "Solar Mini",
"observed": false,
"current": false
},
{
"date": "2024-07-23",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.1,
"valueLabel": "$0.1",
"model": "Llama 3.1 Instruct 8B",
"observed": true,
"current": false
},
{
"date": "2024-11-18",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.088,
"valueLabel": "$0.088",
"model": "Qwen2.5 Turbo",
"observed": true,
"current": false
},
{
"date": "2024-12-03",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.061,
"valueLabel": "$0.061",
"model": "Nova Micro",
"observed": true,
"current": false
},
{
"date": "2026-02-25",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.052,
"valueLabel": "$0.052",
"model": "LFM2 24B A2B",
"observed": true,
"current": false
},
{
"date": "2026-03-02",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.02,
"valueLabel": "$0.02",
"model": "Qwen3.5 0.8B (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-1",
"seriesLabel": "Intelligence 10-20",
"metric": "token",
"panel": "token",
"tierIndex": 1,
"value": 0.02,
"valueLabel": "$0.02",
"model": "Qwen3.5 0.8B (Reasoning)",
"observed": false,
"current": true
},
{
"date": "2024-09-12",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 28.875,
"valueLabel": "$28.9",
"model": "o1-preview",
"observed": true,
"current": false
},
{
"date": "2025-01-31",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 1.925,
"valueLabel": "$1.93",
"model": "o3-mini (high)",
"observed": true,
"current": false
},
{
"date": "2025-03-25",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 1.209,
"valueLabel": "$1.21",
"model": "DeepSeek V3 0324",
"observed": true,
"current": false
},
{
"date": "2025-04-14",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.7,
"valueLabel": "$0.7",
"model": "GPT-4.1 mini",
"observed": true,
"current": false
},
{
"date": "2025-07-21",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.356,
"valueLabel": "$0.356",
"model": "Qwen3 235B A22B 2507 Instruct",
"observed": true,
"current": false
},
{
"date": "2025-07-31",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.352,
"valueLabel": "$0.352",
"model": "Qwen3 Coder 30B A3B Instruct",
"observed": true,
"current": false
},
{
"date": "2025-08-05",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.088,
"valueLabel": "$0.088",
"model": "gpt-oss-20B (high)",
"observed": true,
"current": false
},
{
"date": "2026-03-02",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.06,
"valueLabel": "$0.06",
"model": "Qwen3.5 4B (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-2",
"seriesLabel": "Intelligence 20-30",
"metric": "token",
"panel": "token",
"tierIndex": 2,
"value": 0.06,
"valueLabel": "$0.06",
"model": "Qwen3.5 4B (Non-reasoning)",
"observed": false,
"current": true
},
{
"date": "2024-12-05",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 26.25,
"valueLabel": "$26.3",
"model": "o1",
"observed": true,
"current": false
},
{
"date": "2025-02-19",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.35,
"valueLabel": "$0.35",
"model": "Grok 3 mini Reasoning (high)",
"observed": true,
"current": false
},
{
"date": "2025-08-05",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.262,
"valueLabel": "$0.262",
"model": "gpt-oss-120b (high)",
"observed": true,
"current": false
},
{
"date": "2025-12-16",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.15,
"valueLabel": "$0.15",
"model": "MiMo-V2-Flash (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-03-02",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.113,
"valueLabel": "$0.113",
"model": "Qwen3.5 9B (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-3",
"seriesLabel": "Intelligence 30-40",
"metric": "token",
"panel": "token",
"tierIndex": 3,
"value": 0.113,
"valueLabel": "$0.113",
"model": "Qwen3.5 9B (Reasoning)",
"observed": false,
"current": true
},
{
"date": "2025-06-10",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 35,
"valueLabel": "$35",
"model": "o3-pro",
"observed": true,
"current": false
},
{
"date": "2025-07-10",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 11,
"valueLabel": "$11",
"model": "Grok 4",
"observed": true,
"current": false
},
{
"date": "2025-08-07",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 0.688,
"valueLabel": "$0.688",
"model": "GPT-5 mini (high)",
"observed": true,
"current": false
},
{
"date": "2025-12-01",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 0.337,
"valueLabel": "$0.337",
"model": "DeepSeek V3.2 (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2025-12-16",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 0.15,
"valueLabel": "$0.15",
"model": "MiMo-V2-Flash (Feb 2026)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-4",
"seriesLabel": "Intelligence 40-50",
"metric": "token",
"panel": "token",
"tierIndex": 4,
"value": 0.15,
"valueLabel": "$0.15",
"model": "MiMo-V2-Flash (Feb 2026)",
"observed": false,
"current": true
},
{
"date": "2025-12-11",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 4.813,
"valueLabel": "$4.81",
"model": "GPT-5.2 (xhigh)",
"observed": true,
"current": false
},
{
"date": "2026-02-19",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 4.5,
"valueLabel": "$4.5",
"model": "Gemini 3.1 Pro Preview",
"observed": true,
"current": false
},
{
"date": "2026-04-02",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 1.125,
"valueLabel": "$1.13",
"model": "Qwen3.6 Plus",
"observed": true,
"current": false
},
{
"date": "2026-04-24",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 0.544,
"valueLabel": "$0.544",
"model": "DeepSeek V4 Pro (Reasoning, Max Effort)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "token-5",
"seriesLabel": "Intelligence 50+",
"metric": "token",
"panel": "token",
"tierIndex": 5,
"value": 0.544,
"valueLabel": "$0.544",
"model": "DeepSeek V4 Pro (Reasoning, Max Effort)",
"observed": false,
"current": true
}
]
},
"encoding": {
"x": {
"field": "date",
"type": "temporal",
"scale": {
"domain": [
"2024-03-04",
"2026-05-31"
]
},
"axis": {
"title": null,
"format": "%Y",
"tickCount": {
"interval": "year",
"step": 1
},
"labelFont": "IBM Plex Mono",
"labelFontSize": 11,
"labelColor": "#111111",
"domain": true,
"domainColor": "#d6d6d0",
"domainWidth": 1,
"grid": false
}
},
"y": {
"field": "value",
"type": "quantitative",
"scale": {
"type": "log",
"base": 10,
"domain": [
0.0078125,
64
],
"nice": false
},
"axis": {
"title": "US dollars (log scale)",
"titleFont": "IBM Plex Mono",
"titleFontSize": 14,
"titleColor": "#111111",
"labelFont": "IBM Plex Mono",
"labelFontSize": 11,
"labelColor": "#111111",
"values": [
0.0078125,
0.015625,
0.03125,
0.0625,
0.125,
0.25,
0.5,
1,
2,
4,
8,
16,
32,
64
],
"labelExpr": "datum.value >= 1000 ? '$' + format(datum.value, ',.0f') : datum.value >= 1 ? '$' + format(datum.value, ',.0f') : '$' + format(datum.value, '.4~f')"
}
},
"tooltip": [
{
"field": "date",
"type": "temporal",
"title": "Date"
},
{
"field": "seriesLabel",
"type": "nominal",
"title": "Intelligence band"
},
{
"field": "model",
"type": "nominal",
"title": "Cheapest model"
},
{
"field": "value",
"type": "quantitative",
"title": "List price",
"format": "$,.2f"
}
]
},
"layer": [
{
"mark": {
"type": "line",
"interpolate": "step-after",
"strokeWidth": 1.35,
"strokeCap": "butt",
"color": "#1d1d1b",
"clip": true
},
"encoding": {
"detail": {
"field": "seriesId",
"type": "nominal"
},
"strokeDash": {
"field": "tierIndex",
"type": "ordinal",
"scale": {
"domain": [
0,
1,
2,
3,
4,
5
],
"range": [
[
0.8,
2.2
],
[
2,
2.2
],
[
3.5,
2.4
],
[
5,
2.4
],
[
7,
2.6
],
[
1,
0
]
]
},
"legend": null
}
}
},
{
"transform": [
{
"filter": "datum.observed"
}
],
"mark": {
"type": "point",
"shape": "square",
"size": 34,
"filled": true,
"color": "#1d1d1b",
"clip": true,
"opacity": 1,
"fillOpacity": 1,
"strokeOpacity": 1,
"strokeWidth": 0
}
},
{
"transform": [
{
"filter": "datum.current"
}
],
"mark": {
"type": "point",
"shape": "square",
"size": 42,
"filled": true,
"color": "#1d1d1b",
"clip": false,
"opacity": 1,
"fillOpacity": 1,
"strokeOpacity": 1,
"strokeWidth": 0
}
},
{
"transform": [
{
"filter": "datum.current"
}
],
"mark": {
"type": "text",
"align": "left",
"baseline": "middle",
"dx": 7,
"font": "IBM Plex Mono",
"fontSize": 10,
"fontWeight": 700,
"color": "#111111",
"clip": false
},
"encoding": {
"text": {
"field": "valueLabel",
"type": "nominal"
}
}
}
]
},
{
"title": {
"text": "Cost to finish one standard benchmark task",
"anchor": "start",
"font": "IBM Plex Mono",
"fontSize": 18,
"offset": 12,
"color": "#111111"
},
"width": 470,
"height": 320,
"padding": {
"top": 0,
"right": 44,
"bottom": 0,
"left": 0
},
"data": {
"values": [
{
"date": "2025-03-12",
"seriesId": "aa-0",
"seriesLabel": "Intelligence <10",
"metric": "aa",
"panel": "bill",
"tierIndex": 0,
"value": 6.518,
"valueLabel": "$6.52",
"model": "Gemma 3 4B Instruct",
"observed": true,
"current": false
},
{
"date": "2025-06-26",
"seriesId": "aa-0",
"seriesLabel": "Intelligence <10",
"metric": "aa",
"panel": "bill",
"tierIndex": 0,
"value": 5.7215,
"valueLabel": "$5.72",
"model": "Gemma 3n E4B Instruct",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-0",
"seriesLabel": "Intelligence <10",
"metric": "aa",
"panel": "bill",
"tierIndex": 0,
"value": 5.7215,
"valueLabel": "$5.72",
"model": "Gemma 3n E4B Instruct",
"observed": false,
"current": true
},
{
"date": "2024-03-04",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 36.4377,
"valueLabel": "$36.4",
"model": "Claude 3 Haiku",
"observed": true,
"current": false
},
{
"date": "2024-07-23",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 27.7078,
"valueLabel": "$27.7",
"model": "Llama 3.1 Instruct 8B",
"observed": true,
"current": false
},
{
"date": "2024-12-03",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 6.3848,
"valueLabel": "$6.38",
"model": "Nova Micro",
"observed": true,
"current": false
},
{
"date": "2025-09-22",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 4.4809,
"valueLabel": "$4.48",
"model": "Granite 4.0 H Small",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-1",
"seriesLabel": "Intelligence 10-20",
"metric": "aa",
"panel": "bill",
"tierIndex": 1,
"value": 4.4809,
"valueLabel": "$4.48",
"model": "Granite 4.0 H Small",
"observed": false,
"current": true
},
{
"date": "2025-01-31",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 314.7215,
"valueLabel": "$315",
"model": "o3-mini (high)",
"observed": true,
"current": false
},
{
"date": "2025-03-25",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 59.2474,
"valueLabel": "$59.2",
"model": "DeepSeek V3 0324",
"observed": true,
"current": false
},
{
"date": "2025-04-14",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 53.7334,
"valueLabel": "$53.7",
"model": "GPT-4.1 mini",
"observed": true,
"current": false
},
{
"date": "2025-07-31",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 29.2907,
"valueLabel": "$29.3",
"model": "Qwen3 Coder 30B A3B Instruct",
"observed": true,
"current": false
},
{
"date": "2025-08-05",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 7.6768,
"valueLabel": "$7.68",
"model": "gpt-oss-20B (low)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-2",
"seriesLabel": "Intelligence 20-30",
"metric": "aa",
"panel": "bill",
"tierIndex": 2,
"value": 7.6768,
"valueLabel": "$7.68",
"model": "gpt-oss-20B (low)",
"observed": false,
"current": true
},
{
"date": "2024-12-05",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 3270.3573,
"valueLabel": "$3,270",
"model": "o1",
"observed": true,
"current": false
},
{
"date": "2025-02-19",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 104.8919,
"valueLabel": "$105",
"model": "Grok 3 mini Reasoning (high)",
"observed": true,
"current": false
},
{
"date": "2025-08-05",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 67.3659,
"valueLabel": "$67.4",
"model": "gpt-oss-120b (high)",
"observed": true,
"current": false
},
{
"date": "2025-08-07",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 61.9119,
"valueLabel": "$61.9",
"model": "GPT-5 mini (medium)",
"observed": true,
"current": false
},
{
"date": "2025-09-19",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 35.9855,
"valueLabel": "$36",
"model": "Grok 4 Fast (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2025-12-16",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 21.3815,
"valueLabel": "$21.4",
"model": "MiMo-V2-Flash (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-04-02",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 19.4332,
"valueLabel": "$19.4",
"model": "Gemma 4 31B (Non-reasoning)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-3",
"seriesLabel": "Intelligence 30-40",
"metric": "aa",
"panel": "bill",
"tierIndex": 3,
"value": 19.4332,
"valueLabel": "$19.4",
"model": "Gemma 4 31B (Non-reasoning)",
"observed": false,
"current": true
},
{
"date": "2025-07-10",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 2881.1519,
"valueLabel": "$2,881",
"model": "Grok 4",
"observed": true,
"current": false
},
{
"date": "2025-08-07",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 168.4577,
"valueLabel": "$168",
"model": "GPT-5 mini (high)",
"observed": true,
"current": false
},
{
"date": "2025-12-01",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 75.6823,
"valueLabel": "$75.7",
"model": "DeepSeek V3.2 (Reasoning)",
"observed": true,
"current": false
},
{
"date": "2025-12-16",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 66.5448,
"valueLabel": "$66.5",
"model": "MiMo-V2-Flash (Feb 2026)",
"observed": true,
"current": false
},
{
"date": "2026-04-24",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 57.4318,
"valueLabel": "$57.4",
"model": "DeepSeek V4 Flash (Reasoning, High Effort)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-4",
"seriesLabel": "Intelligence 40-50",
"metric": "aa",
"panel": "bill",
"tierIndex": 4,
"value": 57.4318,
"valueLabel": "$57.4",
"model": "DeepSeek V4 Flash (Reasoning, High Effort)",
"observed": false,
"current": true
},
{
"date": "2025-12-11",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 2304.2197,
"valueLabel": "$2,304",
"model": "GPT-5.2 (xhigh)",
"observed": true,
"current": false
},
{
"date": "2026-02-05",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 1571.5554,
"valueLabel": "$1,572",
"model": "GPT-5.3 Codex (xhigh)",
"observed": true,
"current": false
},
{
"date": "2026-02-19",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 892.2838,
"valueLabel": "$892",
"model": "Gemini 3.1 Pro Preview",
"observed": true,
"current": false
},
{
"date": "2026-04-02",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 482.6505,
"valueLabel": "$483",
"model": "Qwen3.6 Plus",
"observed": true,
"current": false
},
{
"date": "2026-04-22",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 461.5912,
"valueLabel": "$462",
"model": "MiMo-V2.5-Pro",
"observed": true,
"current": false
},
{
"date": "2026-04-24",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 267.8209,
"valueLabel": "$268",
"model": "DeepSeek V4 Pro (Reasoning, Max Effort)",
"observed": true,
"current": false
},
{
"date": "2026-05-31",
"seriesId": "aa-5",
"seriesLabel": "Intelligence 50+",
"metric": "aa",
"panel": "bill",
"tierIndex": 5,
"value": 267.8209,
"valueLabel": "$268",
"model": "DeepSeek V4 Pro (Reasoning, Max Effort)",
"observed": false,
"current": true
}
]
},
"encoding": {
"x": {
"field": "date",
"type": "temporal",
"scale": {
"domain": [
"2024-03-04",
"2026-05-31"
]
},
"axis": {
"title": null,
"format": "%Y",
"tickCount": {
"interval": "year",
"step": 1
},
"labelFont": "IBM Plex Mono",
"labelFontSize": 11,
"labelColor": "#111111",
"domain": true,
"domainColor": "#d6d6d0",
"domainWidth": 1,
"grid": false
}
},
"y": {
"field": "value",
"type": "quantitative",
"scale": {
"type": "log",
"base": 10,
"domain": [
4,
5000
],
"nice": false
},
"axis": {
"title": "US dollars (log scale)",
"titleFont": "IBM Plex Mono",
"titleFontSize": 14,
"titleColor": "#111111",
"labelFont": "IBM Plex Mono",
"labelFontSize": 11,
"labelColor": "#111111",
"values": [
5,
25,
100,
500,
2500,
5000
],
"labelExpr": "datum.value >= 1000 ? '$' + format(datum.value, ',.0f') : datum.value >= 1 ? '$' + format(datum.value, ',.0f') : '$' + format(datum.value, '.4~f')"
}
},
"tooltip": [
{
"field": "date",
"type": "temporal",
"title": "Date"
},
{
"field": "seriesLabel",
"type": "nominal",
"title": "Intelligence band"
},
{
"field": "model",
"type": "nominal",
"title": "Cheapest model"
},
{
"field": "value",
"type": "quantitative",
"title": "AA benchmark bill",
"format": "$,.2f"
}
]
},
"layer": [
{
"mark": {
"type": "line",
"interpolate": "step-after",
"strokeWidth": 1.35,
"strokeCap": "butt",
"color": "#e30022",
"clip": true
},
"encoding": {
"detail": {
"field": "seriesId",
"type": "nominal"
},
"strokeDash": {
"field": "tierIndex",
"type": "ordinal",
"scale": {
"domain": [
0,
1,
2,
3,
4,
5
],
"range": [
[
0.8,
2.2
],
[
2,
2.2
],
[
3.5,
2.4
],
[
5,
2.4
],
[
7,
2.6
],
[
1,
0
]
]
},
"legend": null
}
}
},
{
"transform": [
{
"filter": "datum.observed"
}
],
"mark": {
"type": "point",
"shape": "square",
"size": 34,
"filled": true,
"color": "#e30022",
"clip": true,
"opacity": 1,
"fillOpacity": 1,
"strokeOpacity": 1,
"strokeWidth": 0
}
},
{
"transform": [
{
"filter": "datum.current"
}
],
"mark": {
"type": "point",
"shape": "square",
"size": 42,
"filled": true,
"color": "#e30022",
"clip": false,
"opacity": 1,
"fillOpacity": 1,
"strokeOpacity": 1,
"strokeWidth": 0
}
},
{
"transform": [
{
"filter": "datum.current"
}
],
"mark": {
"type": "text",
"align": "left",
"baseline": "middle",
"dx": 7,
"font": "IBM Plex Mono",
"fontSize": 10,
"fontWeight": 700,
"color": "#111111",
"clip": false
},
"encoding": {
"text": {
"field": "valueLabel",
"type": "nominal"
}
}
}
]
}
],
"spacing": 34,
"align": "all",
"bounds": "full"
},
{
"width": 974,
"height": 64,
"data": {
"values": [
{
"seriesLabel": "Intelligence <10",
"tierIndex": 0,
"xStart": 0,
"xEnd": 0.34,
"labelX": 0.42,
"y": 1
},
{
"seriesLabel": "Intelligence 10-20",
"tierIndex": 1,
"xStart": 1,
"xEnd": 1.34,
"labelX": 1.42,
"y": 1
},
{
"seriesLabel": "Intelligence 20-30",
"tierIndex": 2,
"xStart": 2,
"xEnd": 2.34,
"labelX": 2.42,
"y": 1
},
{
"seriesLabel": "Intelligence 30-40",
"tierIndex": 3,
"xStart": 0,
"xEnd": 0.34,
"labelX": 0.42,
"y": 0
},
{
"seriesLabel": "Intelligence 40-50",
"tierIndex": 4,
"xStart": 1,
"xEnd": 1.34,
"labelX": 1.42,
"y": 0
},
{
"seriesLabel": "Intelligence 50+",
"tierIndex": 5,
"xStart": 2,
"xEnd": 2.34,
"labelX": 2.42,
"y": 0
}
]
},
"title": {
"text": "Intelligence band",
"anchor": "start",
"font": "IBM Plex Mono",
"fontSize": 15,
"color": "#111111"
},
"layer": [
{
"mark": {
"type": "rule",
"strokeWidth": 1.35,
"color": "#111111"
},
"encoding": {
"x": {
"field": "xStart",
"type": "quantitative",
"axis": null,
"scale": {
"domain": [
0,
3
]
}
},
"x2": {
"field": "xEnd"
},
"y": {
"field": "y",
"type": "quantitative",
"axis": null,
"scale": {
"domain": [
-0.35,
1.35
]
}
},
"strokeDash": {
"field": "seriesLabel",
"type": "nominal",
"scale": {
"domain": [
"Intelligence <10",
"Intelligence 10-20",
"Intelligence 20-30",
"Intelligence 30-40",
"Intelligence 40-50",
"Intelligence 50+"
],
"range": [
[
0.8,
2.2
],
[
2,
2.2
],
[
3.5,
2.4
],
[
5,
2.4
],
[
7,
2.6
],
[
1,
0
]
]
},
"legend": null
}
}
},
{
"mark": {
"type": "text",
"align": "left",
"baseline": "middle",
"font": "IBM Plex Mono",
"fontSize": 15,
"color": "#111111"
},
"encoding": {
"x": {
"field": "labelX",
"type": "quantitative",
"axis": null,
"scale": {
"domain": [
0,
3
]
}
},
"y": {
"field": "y",
"type": "quantitative",
"axis": null,
"scale": {
"domain": [
-0.35,
1.35
]
}
},
"text": {
"field": "seriesLabel",
"type": "nominal"
}
}
}
]
}
],
"spacing": 8,
"resolve": {
"scale": {
"x": "shared",
"y": "independent",
"color": "independent",
"strokeDash": "independent"
}
}
},
"notes": "Two ways to read model cost. Left: the sticker price, what the provider charges per million tokens blended at the standard 3:1 input-to-output ratio. Right: the bill for running the full Artificial Analysis Intelligence Index on that model. Data checked on 31 May 2026. Across the six bands shown, the median cheapest token price has fallen about 120x since March 2024; the median benchmark bill has fallen about 25x. The two prices come apart because models use different amounts of input, reasoning and answer tokens. Cheap tokens do not always mean a cheap finished job, and pricey tokens do not always mean a pricey one. Each line tracks the cheapest model available in a given intelligence band; lower is cheaper. Data from Artificial Analysis [E210]; intelligence bands after a16z [E178]."
},
{
"id": "model-economics-lane-bridge",
"evidenceIds": [
"E20",
"E21"
],
"type": "paragraph",
"text": "Sticker price is only the first cost signal. Lane choice determines whether the same workload is interactive, batched, cached, flexed, provisioned or delayed, which changes both the bill and the review burden.",
"variant": "body"
},
{
"id": "model-pricing-mechanics-async-lanes-lead",
"evidenceIds": [],
"type": "paragraph",
"text": "The best discount for filing corpora, KYC file drops, claims batches, policy archives and private eval reruns is often patience. Long extraction packets, eval reruns, classification sweeps and archive enrichment do not need a user waiting in the browser. Move that work to async or lower-priority lanes, keep deadline work on the fast path, and compare cost per financial output.",
"variant": "body"
},
{
"id": "model-pricing-mechanics-pricing-lanes-table",
"title": "The discount lane is part of the route",
"evidenceIds": [],
"type": "table",
"columns": [
{
"key": "lane",
"label": "Lane",
"align": "left"
},
{
"key": "how_it_works",
"label": "How it works",
"align": "left"
},
{
"key": "good_for",
"label": "Good for",
"align": "left"
},
{
"key": "watch",
"label": "Watch",
"align": "left"
}
],
"rows": [
{
"lane": "OpenAI Batch API",
"how_it_works": "Async jobs use a separate batch lane with a lower price and a 24-hour completion target. [E185]",
"good_for": "Large document extraction, eval reruns, classification, enrichment and non-urgent archive work.",
"watch": "Not an interactive path; queue time, file packaging, partial failures and output reconciliation become part of the route."
},
{
"lane": "OpenAI Flex processing",
"how_it_works": "Eligible calls use lower-priority capacity and are charged at Batch API rates for supported models. [E195]",
"good_for": "Long-document async work where cost matters more than immediate availability.",
"watch": "Treat it as a cheaper lane with lower priority, not as the path for a signing call, filing deadline or live customer answer."
},
{
"lane": "Anthropic Message Batches",
"how_it_works": "Bulk async requests can be priced below the standard synchronous lane and can be combined with prompt caching. [E197, E198]",
"good_for": "Large non-interactive packs: extraction, summarization, evals, source tagging and overnight review queues.",
"watch": "Batch queue behaviour and cache checkpoint design decide whether the savings survive production."
},
{
"lane": "Gemini / Vertex AI batch",
"how_it_works": "Vertex batch prediction is the non-real-time lane; Google pricing separates standard, batch and cached-input treatment. [E199, E201]",
"good_for": "Bulk document analysis, classification and back-office model calls that do not need immediate response.",
"watch": "Batch mode, provisioned throughput and cached context are separate mechanisms; do not assume one discount buys the others."
},
{
"lane": "Amazon Bedrock batch inference",
"how_it_works": "Bedrock batch inference runs async jobs over S3 inputs, with discounted batch pricing for supported models. [E202, E204]",
"good_for": "Cloud-controlled bulk inference where the input/output artefacts can live in the approved AWS path.",
"watch": "Prompt caching and batch inference are not universally combinable; model, region and feature support decide the real lane."
},
{
"lane": "Groq Batch and Flex",
"how_it_works": "Groq's Batch API is a discounted async lane; Flex processing trades immediacy for access to available capacity rather than a lower price. [E206, E207]",
"good_for": "Throughput-sensitive work where the buyer can separate background processing from live assistant use.",
"watch": "Low published developer limits can still make a clever route unusable until the account, plan or enterprise path changes."
},
{
"lane": "Usage tiers and quota requests",
"how_it_works": "Spend tiers, usage tiers, quota requests and account-specific limits can change the ceiling before any model changes. [E184, E186, E187, E188, E190, E192]",
"good_for": "New projects that need to absorb a sudden file drop, launch backlog or review queue.",
"watch": "Tier language is not capacity evidence. Check the live project, account, subscription, region and deployment type."
},
{
"lane": "Provisioned or reserved capacity",
"how_it_works": "Azure PTUs and reservations, Vertex Provisioned Throughput and Bedrock Provisioned Throughput buy predictability rather than a smarter model. [E189, E191, E194]",
"good_for": "Recurring volume, deadline-sensitive work and routes where shared-capacity 429s would be business failures.",
"watch": "Commitments, idle capacity, regional availability, spillover rules and reservation terms belong in the price model."
}
]
},
{
"id": "model-pricing-mechanics-cache-lead",
"evidenceIds": [],
"type": "paragraph",
"text": "Prompt caching is easy to misunderstand. It is not model memory and it does not cache the answer. It reuses the provider's work from the repeated beginning of a prompt: instructions, tool schemas, output schema, playbook text, matter taxonomy or source bundle. If the next request starts the same way, the provider can bill and process that repeated prefix differently. If the route prepends a timestamp, request id or one-off user sentence before the stable material, the cache may miss.",
"variant": "body"
},
{
"id": "model-pricing-mechanics-cache-design-lead",
"evidenceIds": [],
"type": "paragraph",
"text": "Design the prompt like a production artefact. Put the durable, reusable context first; put the unique question last; record cache hits in the financial model route register. Cache savings are real only when the route keeps a stable prefix and still produces financial output.",
"variant": "body"
},
{
"id": "model-pricing-mechanics-cache-design-table",
"title": "How to make caching work",
"evidenceIds": [],
"type": "table",
"columns": [
{
"key": "design_rule",
"label": "Design rule",
"align": "left"
},
{
"key": "why_it_matters",
"label": "Why it matters",
"align": "left"
},
{
"key": "implementation",
"label": "Implementation",
"align": "left"
}
],
"rows": [
{
"design_rule": "Put stable material first",
"why_it_matters": "Most cache mechanisms reward an identical or explicitly marked prefix.",
"implementation": "Start with system instructions, tool schemas, output schema, jurisdiction rules, playbook text and reusable source packs."
},
{
"design_rule": "Put variable material last",
"why_it_matters": "A unique timestamp, request id, user aside or matter-specific fact at the top can break a prefix match.",
"implementation": "Append the specific question, run id, new facts, reviewer note and requested output after the cacheable prefix."
},
{
"design_rule": "Keep tools and schemas stable",
"why_it_matters": "Tool definitions, JSON schemas and message ordering can be part of the cache key.",
"implementation": "Version schemas deliberately; avoid rebuilding tool lists, whitespace, examples or column order on every call."
},
{
"design_rule": "Warm the cache before fan-out",
"why_it_matters": "Parallel extraction can waste money if every worker pays to prefill the same long context.",
"implementation": "Send one cacheable request for the source pack or playbook, then fan out clause, entity, citation or issue tasks."
},
{
"design_rule": "Cache source packs, not shortcuts",
"why_it_matters": "The cache should preserve approved context, not hide missing retrieval or stale policy.",
"implementation": "Attach source version, route version and expiry; rebuild the prefix when authority, playbook or source set changes."
},
{
"design_rule": "Measure cached-token economics",
"why_it_matters": "A cache discount that never hits is procurement fiction.",
"implementation": "Log cache reads, writes, cached tokens, hit rate, latency and accepted-output rate beside the normal token bill."
},
{
"design_rule": "Respect the data boundary",
"why_it_matters": "Caching is still provider processing under the approved route.",
"implementation": "Cache only context that is allowed inside the approved data boundary. Caching is a billing and latency control, not permission to reuse one client's facts in another client's work."
}
]
},
{
"id": "model-pricing-mechanics-provider-cache-table",
"title": "Provider caching is not one thing",
"evidenceIds": [],
"type": "table",
"columns": [
{
"key": "provider",
"label": "Provider",
"align": "left"
},
{
"key": "mechanic",
"label": "Mechanic",
"align": "left"
},
{
"key": "buyer_read",
"label": "Buyer read",
"align": "left"
}
],
"rows": [
{
"provider": "OpenAI",
"mechanic": "Automatic prompt caching applies to repeated prompt prefixes above the documented threshold; usage surfaces cached tokens, and prompt-cache keys can help route related work. [E196]",
"buyer_read": "Keep the first part of the prompt identical: instructions, tools, schema and reusable source context before matter-specific text."
},
{
"provider": "Anthropic",
"mechanic": "Claude uses explicit cache-control breakpoints over tools, system and message content, with separate cache write and read economics and documented cache durations. [E198]",
"buyer_read": "Design cache checkpoints deliberately. The best target is the long shared context, not the short one-off question."
},
{
"provider": "Gemini / Vertex AI",
"mechanic": "Vertex supports context caching, including explicit cached content and implicit prefix caching; pricing separates cached input from ordinary input and batch treatment. [E199, E200]",
"buyer_read": "Check whether the route uses explicit cache objects, implicit prefix reuse or batch prediction; each has different cost and expiry behaviour."
},
{
"provider": "Amazon Bedrock",
"mechanic": "Bedrock prompt caching uses cache points with time-to-live controls and separate cache pricing, while batch inference has its own feature constraints. [E202, E203, E204]",
"buyer_read": "Do not assume a cached interactive prompt can simply move into batch. Confirm feature support by model and region."
},
{
"provider": "Groq",
"mechanic": "Groq prompt caching is automatic for exact repeated prefixes above the documented threshold, with discounted cached tokens and separate treatment in rate-limit accounting. [E205]",
"buyer_read": "The route must keep a stable prefix. This matters most when rate limits are low and large token packets would otherwise queue."
}
]
},
{
"id": "model-pricing-mechanics-approval-lead",
"evidenceIds": [],
"type": "paragraph",
"text": "The practical question is whether the buyer can deliberately choose the cheap lane without breaking the work. A cheaper async run is useful for overnight extraction. It is dangerous for an answer needed during a signing call. A cache hit is useful when the prefix is stable. It is imaginary when every prompt is rebuilt from scratch.",
"variant": "body"
},
{
"id": "model-pricing-mechanics-approval-table",
"title": "What to prove before calling the route cheap",
"evidenceIds": [],
"type": "table",
"columns": [
{
"key": "test",
"label": "Test",
"align": "left"
},
{
"key": "proof",
"label": "Proof",
"align": "left"
},
{
"key": "pass",
"label": "Pass condition",
"align": "left"
}
],
"rows": [
{
"test": "Interactive versus async split",
"proof": "Tag each workflow step as user-blocking, same-day, overnight or archive work.",
"pass": "Only non-blocking work moves to batch, flex or lower-priority processing."
},
{
"test": "Batch economics",
"proof": "Run the same packet through interactive and batch lanes, including failed files, retries, queue time and output handling.",
"pass": "The cheaper lane still produces accepted financial output inside the work window."
},
{
"test": "Cache hit rate",
"proof": "Log cache reads, writes, cached tokens or provider-equivalent fields by route version.",
"pass": "The actual hit rate is high enough to change the finished-task bill, not just the theoretical token price."
},
{
"test": "Tier and capacity path",
"proof": "Capture account tier, quota, regional capacity, provisioned-capacity option and support escalation.",
"pass": "The buyer knows which spend, quota or reserved-capacity lever unlocks the next workload level."
},
{
"test": "Accepted-output denominator",
"proof": "Combine model spend, queue time, cache hit rate, reviewer minutes, failed runs and rework.",
"pass": "The route is cheaper per financial output, not merely cheaper per million tokens."
}
]
},
{
"id": "accepted-output-cost-table",
"title": "Cost per accepted output",
"evidenceIds": [
"E20",
"E21"
],
"type": "table",
"columns": [
{
"key": "cost_item",
"label": "Cost item",
"align": "left"
},
{
"key": "token_view",
"label": "Token view",
"align": "left"
},
{
"key": "accepted_output_view",
"label": "Accepted-output view",
"align": "left"
},
{
"key": "measurement",
"label": "Measurement",
"align": "left"
}
],
"rows": [
{
"cost_item": "Retrieval miss",
"token_view": "Cheap answer.",
"accepted_output_view": "Expensive wrong answer.",
"measurement": "Correct-source recall and no-answer rate."
},
{
"cost_item": "Table loss",
"token_view": "Normal prompt.",
"accepted_output_view": "Review fails because the key row vanished.",
"measurement": "Table-cell preservation and source-span export."
},
{
"cost_item": "Calculation error",
"token_view": "Low inference cost.",
"accepted_output_view": "Analyst recalculates or rejects.",
"measurement": "Formula accuracy, intermediate values and reviewer minutes."
},
{
"cost_item": "Long context",
"token_view": "More tokens.",
"accepted_output_view": "Better evidence, higher latency and higher cost.",
"measurement": "Accepted answer rate per minute and dollar."
},
{
"cost_item": "Productivity claim",
"token_view": "Low model bill.",
"accepted_output_view": "Savings are real only when accepted output, reviewer time or revenue lift is measured.",
"measurement": "Accepted outputs per reviewer hour, baseline labour cost and revenue-attribution confidence."
},
{
"cost_item": "Refusal",
"token_view": "No hallucination.",
"accepted_output_view": "Good if evidence is missing; broken if evidence exists.",
"measurement": "Correct refusal, false refusal and missing-source explanation."
},
{
"cost_item": "Audit export",
"token_view": "Not counted.",
"accepted_output_view": "Required for exams, disputes and model validation.",
"measurement": "Reproducible inputs, source IDs, model version and reviewer decision."
}
]
},
{
"id": "route-economics-table",
"title": "How to price an accepted output",
"evidenceIds": [
"E20",
"E21"
],
"type": "table",
"columns": [
{
"key": "workflow",
"label": "Workflow",
"align": "left"
},
{
"key": "why_token_price_misleads",
"label": "Why token price misleads",
"align": "left"
},
{
"key": "buyer_test",
"label": "Buyer test",
"align": "left"
}
],
"rows": [
{
"workflow": "Public filing Q&A",
"why_token_price_misleads": "The answer lives in tables, notes, periods and restatements; a cheap answer is expensive if the source is wrong.",
"buyer_test": "Measure correct-source recall, source-span export, no-answer behaviour and reviewer minutes."
},
{
"workflow": "Policy search",
"why_token_price_misleads": "The model bill is small; stale retrieval, missing policy exceptions and archive gaps create the real cost.",
"buyer_test": "Track current-source recall, correct refusal and audit-export completeness."
},
{
"workflow": "Finance productivity route",
"why_token_price_misleads": "AI productivity gains may appear before revenue attribution; seats, usage and compute charges can all misstate the accepted-output economics.",
"buyer_test": "Measure accepted outputs per reviewer hour, backfill avoided, reruns and confidence in revenue attribution."
},
{
"workflow": "Credit decision support",
"why_token_price_misleads": "Customer outcome, adverse-action reasons and fairness checks move cost into validation and review.",
"buyer_test": "Measure reason-code accuracy, prohibited-proxy checks, sign-off time and archive record quality."
},
{
"workflow": "AML case support",
"why_token_price_misleads": "False negatives, entity linkage and investigation history matter more than fluent summaries.",
"buyer_test": "Measure missed-link rate, investigator disposition, escalation quality and reconstructed evidence packet."
}
]
},
{
"id": "model-pricing-mechanics-rule",
"title": "Pricing mechanics rule",
"evidenceIds": [],
"type": "callout",
"tone": "info",
"body": "Do not compare only the model row. First choose the lane: interactive, batch, lower-priority or flex, cached, provisioned, reserved or self-hosted. The same model can have different cost, latency, quota and review economics depending on the lane."
},
{
"id": "model-economics-section-callout",
"title": "Accepted output is the unit",
"evidenceIds": [
"E20",
"E21"
],
"type": "callout",
"tone": "info",
"body": "A cheap model becomes expensive when it adds review minutes, missed evidence, failed reconciliations or repeat runs."
}
]
},
{
"id": "throughput-rate-limits",
"title": "Throughput and rate limits",
"access": "locked",
"dek": "Rate limits decide whether the model can survive a live financial workload.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "throughput-rate-limits-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "model-lanes",
"title": "Model lanes",
"access": "locked",
"dek": "The finance shortlist should mix generators, retrievers, parsers and private models.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "model-lanes-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "use-case-routing",
"title": "Use-case route selection",
"access": "locked",
"dek": "The same model should not receive the same powers everywhere.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "use-case-routing-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "benchmarks",
"title": "Benchmarks",
"access": "locked",
"dek": "Public scores show where to test, not what to buy.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "benchmarks-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "source-pipeline",
"title": "Source evidence path",
"access": "locked",
"dek": "Financial evidence is structured, temporal and permissioned.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "source-pipeline-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "regulatory-gates",
"title": "Regulatory gates",
"access": "locked",
"dek": "Model approval follows intended use, jurisdiction and customer impact.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "regulatory-gates-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "deployment-routes",
"title": "Deployment controls",
"access": "locked",
"dek": "The deployment path decides what the model may see and what the institution can prove.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "deployment-routes-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "agent-controls",
"title": "Agent controls",
"access": "locked",
"dek": "Finance agents need action governance before autonomy.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "agent-controls-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "eval-design",
"title": "Eval design",
"access": "locked",
"dek": "The private eval is the procurement brief.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "eval-design-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "implementation-plan",
"title": "Implementation plan",
"access": "locked",
"dek": "Build the programme around control paths, not pilots.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "implementation-plan-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "risk-controls",
"title": "Risk controls",
"access": "locked",
"dek": "Finance AI needs route controls, not a model-use policy alone.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "risk-controls-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "cross-border-limits",
"title": "Cross-border limits",
"access": "locked",
"dek": "Global finance breaks simple model approval.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "cross-border-limits-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "model-shortlist",
"title": "Model shortlist",
"access": "locked",
"dek": "Shortlist model lanes by workflow, not by leaderboard.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "model-shortlist-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "implementation-cases",
"title": "Implementation cases",
"access": "locked",
"dek": "The strongest cases sell the operating wrapper, not the model alone.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "implementation-cases-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "operational-signals",
"title": "Operational signals",
"access": "locked",
"dek": "Model governance needs monitoring outside the vendor deck.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "operational-signals-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "expert-watchlist",
"title": "Expert watchlist",
"access": "locked",
"dek": "Watch people who publish useful tests, failures and boundary conditions.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "expert-watchlist-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "vendor-directory",
"title": "Vendor directory",
"access": "locked",
"dek": "Use the directory to route diligence by layer.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "vendor-directory-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "known-unknowns",
"title": "Known unknowns",
"access": "locked",
"dek": "Missing evidence should become a purchasing test.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "known-unknowns-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "source-ledger",
"title": "Source ledger",
"access": "locked",
"dek": "Primary sources, benchmarks and regulatory anchors.",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "source-ledger-locked",
"title": "Locked section",
"evidenceIds": [],
"type": "callout",
"tone": "locked",
"body": "Purchase or subscribe to unlock this section."
}
]
},
{
"id": "glossary",
"title": "Glossary",
"access": "sample",
"claimIds": [],
"evidenceIds": [],
"blocks": [
{
"id": "glossary-table",
"title": "Terms used in this report",
"caption": "Definitions are reader aids, not separate evidence claims.",
"evidenceIds": [],
"type": "table",
"columns": [
{
"key": "term",
"label": "Term",
"align": "left"
},
{
"key": "meaning",
"label": "Plain meaning",
"align": "left"
}
],
"rows": [
{
"term": "Route",
"meaning": "The approved path from work to accepted output: source set, data boundary, model or vendor channel, tools, reviewer, evidence record and reapproval trigger."
},
{
"term": "API",
"meaning": "Application programming interface: a governed software connection used by an app or workflow to call a model or service."
},
{
"term": "SDK",
"meaning": "Software development kit: packaged code and tools developers use to connect a product to an API or service."
},
{
"term": "LLM",
"meaning": "Large language model: a model trained to read and generate text-like tokens."
},
{
"term": "Token",
"meaning": "A chunk of text used for billing, context windows and rate limits. It is usually smaller than a word, not a page count."
},
{
"term": "RPM",
"meaning": "Requests per minute: how many separate API calls the route can accept each minute."
},
{
"term": "RPD",
"meaning": "Requests per day: the daily cap on separate API calls."
},
{
"term": "TPM",
"meaning": "Tokens per minute: how much token traffic a route can process each minute, unless the provider splits input and output separately."
},
{
"term": "TPD",
"meaning": "Tokens per day: the daily cap on token traffic."
},
{
"term": "ITPM",
"meaning": "Input tokens per minute: how many prompt and source-document tokens may enter the model each minute."
},
{
"term": "OTPM",
"meaning": "Output tokens per minute: how many generated tokens may leave the model each minute."
},
{
"term": "PTU",
"meaning": "Provisioned throughput unit: reserved Azure OpenAI capacity bought for steadier throughput."
},
{
"term": "GSU",
"meaning": "Google scale unit: provisioned Vertex AI capacity used to reserve Gemini throughput."
},
{
"term": "MU",
"meaning": "Model Unit: Amazon Bedrock provisioned-throughput capacity reserved for a specific model route."
},
{
"term": "PayGo",
"meaning": "Pay as you go: shared cloud capacity billed by usage rather than pre-reserved throughput."
},
{
"term": "RAG",
"meaning": "Retrieval-augmented generation: search approved sources first, then pass selected evidence into the model."
},
{
"term": "OCR",
"meaning": "Optical character recognition: converting scanned or image text into machine-readable text."
},
{
"term": "SLA",
"meaning": "Service-level agreement: a contractual availability or support commitment, not observed uptime history."
},
{
"term": "JSON",
"meaning": "JavaScript Object Notation: a structured text format commonly used for machine-readable API output."
},
{
"term": "429",
"meaning": "HTTP 429 means too many requests: the route hit a rate limit or shared-capacity throttle."
}
]
}
]
}
]
}
},
"gap_analysis": "Comparable public benchmarks, downstream parser-to-answer tests, longitudinal product telemetry, permissioned-corpus datasets, AML and fraud false-negative datasets, customer-outcome studies, model-change regression logs, concentration-event data and audit-export evidence remain missing for financial model routes.",
"kinetics": {
"speed": "fast",
"impact": 0.89,
"momentum": 0.86,
"businessImpact": "Finance model approval has shifted from choosing an LLM to proving the evidence route can survive review.",
"rateOfChange": "AI Act consultations, GPAI enforcement and finance benchmark refreshes can change credit, insurance, filing-RAG and agent routes before renewal.",
"from": "general chat pilots",
"toward": "routed finance stacks with source controls, document intelligence, model-risk evidence, tool gates and human review"
},
"watch_signals": [
"23 June 2026: European Commission high-risk AI consultation closes; credit and insurance route classification may move [E180].",
"2 August 2026: Commission GPAI enforcement powers begin; model documentation, transparency and provider-diligence evidence may change [E181]."
],
"sample_findings": [
"The best financial AI strategy is an approved-output portfolio, not a single-model mandate."
]
}
```