Best AI models for finance
Rankings by use case and by benchmark. There is no combined score: every ranking names the benchmark it comes from and when the number was reported.
Use cases8 kinds of finance task
- Best AI for Analyst researchFrom Deep FinResearch Bench, Finance Agent Benchmark, FinDeepResearch, FinResearchBench, FrontierFinance, Vals Finance Index
- Best AI for Filings Q&AFrom CorpFin v2, FinanceBench, FinanceQA, IPO Finance Agent
- Best AI for Live market dataFrom CAIA (Crypto AI Agent Benchmark), FinSearchComp
- Best AI for Finance knowledgeFrom Advanced Financial Reasoning at Scale (CFA Level III evaluation), BizFinBench, Open FinLLM Leaderboard, Reasoning Models Ace the CFA Exams (CFA mock-exam evaluation), Tax Agent Bench
- Best AI for VerifiabilityFrom ClaimBench, FAITH, FinLFQA, K-FinHallu
- Best AI for Advice and safetyFrom FinSafetyBench, FinTrust
- Best AI for Trading and portfoliosFrom Agent Market Arena (AMA), AI-Trader, InvestorBench, KTD-Fin, LiveTradeBench, PolyBench, PortBench, StockBench
- Best AI for Spreadsheets and accountingFrom APEX-Accounting, BlueFin, Excel Modeling Benchmark (EMB), Finch: Finance & Accounting Cowork Benchmark
By benchmark38 benchmarks with scores
- AA-AnalystAgentSpreadsheets and accounting · Gemini 3.7 Flash (High) leads, 60 pass^5 %
- Advanced Financial Reasoning at Scale (CFA Level III evaluation)Finance knowledge · o4-mini leads, 79.1 overall % (mean of MCQ and essay)
- Agent Market Arena (AMA)Trading and portfolios · InvestorAgent / GPT-4.1 leads, 40.8 cumulative return % (TSLA)
- AI-TraderTrading and portfolios · MiniMax-M2 leads, 9.6 cumulative return %
- APEX-AccountingSpreadsheets and accounting · Gemini 4 Argon leads, 71.5 Mean Score (% criteria passed)
- BizFinBenchFinance knowledge · ChatGPT-o3 leads, 73.9 average score across 9 tasks
- CAIA (Crypto AI Agent Benchmark)Live market data · gpt-5 leads, 67.4 accuracy % (majority vote, 5 runs)
- ClaimBenchVerifiability · d1-ultra leads, 100 accuracy %
- CorpFin v2Filings Q&A · Claude Opus 5 leads, 73.2 accuracy %
- Deep FinResearch BenchAnalyst research · OpenAI leads, 86 overall factual claim rate %
- Excel Modeling Benchmark (EMB)Spreadsheets and accounting · Claude Fable 5.1 leads, 76.7 accuracy %
- FAITHVerifiability · Claude Sonnet 4 leads, 95.6 overall accuracy %
- Finance Agent BenchmarkAnalyst research · Claude Opus 5.5 leads, 67.8 Partial Credit accuracy %
- FinanceBenchFilings Q&A · GPT-4-Turbo leads, 85 % correct (150-question sample)
- Finch: Finance & Accounting Cowork BenchmarkSpreadsheets and accounting · GPT 5.1 Pro leads, 38.4 workflow pass rate % (human eval)
- FinDeepResearchAnalyst research · FinAgentTest leads, 41.9 overall accuracy score (0-100)
- FinResearchBenchAnalyst research · Gemini leads, 76.6 Final Score (0-100)
- FinSafetyBenchAdvice and safety · LLaMA-3 leads, 32.3 attack success rate % (financial crimes, English)
- FinSearchCompLive market data · grok-4 leads, 68.9 Global subset average accuracy %
- FrontierFinanceAnalyst research · Samaya Sterling leads, 58.9 rubrics qualification rate %
- InvestorBenchTrading and portfolios · Qwen2.5-72B-Instruct leads, 46.2 cumulative return %
- IPO Finance AgentFilings Q&A · GLM-5.2 leads, 79.8 mean rubric score %
- K-FinHalluVerifiability · Gemini 2.5 Flash leads, 0.860 binary detection F1
- KTD-FinTrading and portfolios · Qwen3.6-Plus leads, 85.3 total return %
- LiveTradeBenchTrading and portfolios · GPT-4.1 leads, 6.3 cumulative return %
- Market-BenchTrading and portfolios · grok-4 leads, 443.2 mean MAE
- Open FinLLM LeaderboardFinance knowledge · GPT-4o leads, 46.0 balanced score (mean of text/vision/audio averages)
- PolyBenchTrading and portfolios · MiMo-V2-Flash leads, 17.6 Confidence-Weighted Return (CWR) %
- PRBench FinanceAdvice and safety · Muse Spark 1.3 leads, 59.5 weighted rubric score (0-100)
- Reasoning Models Ace the CFA Exams (CFA mock-exam evaluation)Finance knowledge · Gemini 3.0 Pro leads, 97.6 Level I accuracy %
- StockBenchTrading and portfolios · Qwen3-235B-Think leads, 2.5 final return %
- Tax Agent BenchFinance knowledge · Qwen 3.8 Max 0902 leads, 49.7 accuracy % (all-pass x citation factor)
- Vals Finance IndexAnalyst research · Gemini 4 Carbon leads, 66.3 index accuracy %