FinQA (Snorkel OpenEnv environment)
Agent answers 290 multi-step questions on SEC 10-K data via MCP tools and constrained SQL. Binary reward with 1% numeric tolerance.
Not the 2021 FinQA dataset. Design choices: SQL tool rejects SELECT * and requires filters. No LLM judge. 50-tool-call cap. An RL-trained Qwen3-4B (59.7%) beat Qwen3-235B (51.4%), which Snorkel reads as tool discipline over scale. Also hosted as HF Space snorkelai/finqa-env.
- Made by
- Snorkel AI
- Released
- March 2026
- Kind
- Finance
- Level
- University
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Strongest model tested
-
Gemini 3 Pro #55 on the capability index - Last model added
- November 2025
- Scores refreshed
- November 12th, 2026
2 models share the top score of 81% among models released by November 2025
Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
The newest model with a score here came out in November 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
- Gemini 4 Carbon
- GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 81.0 o3
- 81.0 GPT-5
- 80.3 Gemini 3 Pro
- 80.3 Claude Opus 4.1
- 79.3 GPT-5 mini
- 78.3 Claude Opus 4
- 77.9 Claude 3.7 Sonnet
- 76.6 o4-mini
- 76.6 Claude Sonnet 4
- 74.0 Grok 4
- 73.5 Grok 4 Fast
- 71.7 Kimi K2 Thinking
- 66.6 gpt-oss-120b
- 65.9 Grok 3
- 63.8 o3-mini
- 62.7 GPT-4.1
- 62.1
Amazon Nova Premier - 60.6 Gemini 2.5 Pro [snapshot unspecified]
- 53.1 Gemini 2.5 Flash [snapshot unspecified]
- 51.4 Qwen 3 235B [snapshot unspecified]
- 50.0 GPT-5 nano
- 44.0
Llama-3.3-Nemotron-Super-49B-v1.5 - 40.3 Amazon Nova Pro
- 27.6 Codestral [snapshot unspecified]
- 16.9 Amazon Nova Lite
- 16.2 Magistral Medium [snapshot unspecified]
- 14.5 Amazon Nova Micro
- 13.4 Mistral Large [snapshot unspecified]
- 1 o3 81%
- 2 GPT-5 81%
- 3 Gemini 3 Pro 80.3%
- 4 Claude Opus 4.1 80.3%
- 5 GPT-5 mini 79.3%
- 6 Claude Opus 4 78.3%
- 7 Claude 3.7 Sonnet 77.9%
- 8 o4-mini 76.6%
- 9 Claude Sonnet 4 76.6%
- 10 Grok 4 74%
- 11 Grok 4 Fast reasoning 73.5%
- 12 Kimi K2 Thinking 71.7%
- 13 gpt-oss-120b 66.6%
- 14 Grok 3 65.9%
- 15 o3-mini 63.8%
- 16 GPT-4.1 62.7%
- 17 Amazon Nova Premier 62.1%
- 18 Gemini 2.5 Pro [snapshot unspecified] 60.6%
- 19 Gemini 2.5 Flash [snapshot unspecified] 53.1%
- 20 Qwen 3 235B [snapshot unspecified] 51.4%
- 21 GPT-5 nano 50%
- 22 Llama-3.3-Nemotron-Super-49B-v1.5 44%
- 23 Amazon Nova Pro 40.3%
- 24 Codestral [snapshot unspecified] 27.6%
- 25 Amazon Nova Lite 16.9%
- 26 Magistral Medium [snapshot unspecified] 16.2%
- 27 Amazon Nova Micro 14.5%
- 28 Mistral Large [snapshot unspecified] 13.4%
Source: Official FinQA (Snorkel OpenEnv environment) leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | o3 | 81% | Apr 2025 | Official leaderboard | |
| 2 | GPT-5 | 81% | Aug 2025 | Official leaderboard | |
| 3 | Gemini 3 Pro | 80.3% | Nov 2025 | Official leaderboard | |
| 4 | Claude Opus 4.1 | 80.3% | Aug 2025 | Official leaderboard | |
| 5 | GPT-5 mini | 79.3% | Aug 2025 | Official leaderboard | |
| 6 | Claude Opus 4 | 78.3% | May 2025 | Official leaderboard | |
| 7 | Claude 3.7 Sonnet | 77.9% | Feb 2025 | Official leaderboard | |
| 8 | o4-mini | 76.6% | Apr 2025 | Official leaderboard | |
| 9 | Claude Sonnet 4 | 76.6% | May 2025 | Official leaderboard | |
| 10 | Grok 4 | 74% | Jul 2025 | Official leaderboard | |
| 11 | Grok 4 Fast | 73.5% | reasoning | Sep 2025 | Official leaderboard |
| 12 | Kimi K2 Thinking | 71.7% | Nov 2025 | Official leaderboard | |
| 13 | gpt-oss-120b | 66.6% | Aug 2025 | Official leaderboard | |
| 14 | Grok 3 | 65.9% | Apr 2025 | Official leaderboard | |
| 15 | o3-mini | 63.8% | Jan 2025 | Official leaderboard | |
| 16 | GPT-4.1 | 62.7% | Apr 2025 | Official leaderboard | |
| 17 | Amazon Nova Premier | 62.1% | Apr 2025 | Official leaderboard | |
| 18 | Gemini 2.5 Pro [snapshot unspecified] | 60.6% | Official leaderboard | ||
| 19 | Gemini 2.5 Flash [snapshot unspecified] | 53.1% | Official leaderboard | ||
| 20 | Qwen 3 235B [snapshot unspecified] | 51.4% | Official leaderboard | ||
| 21 | GPT-5 nano | 50% | Aug 2025 | Official leaderboard | |
| 22 | Llama-3.3-Nemotron-Super-49B-v1.5 | 44% | Jul 2025 | Official leaderboard | |
| 23 | Amazon Nova Pro | 40.3% | Dec 2024 | Official leaderboard | |
| 24 | Codestral [snapshot unspecified] | 27.6% | Official leaderboard | ||
| 25 | Amazon Nova Lite | 16.9% | Dec 2024 | Official leaderboard | |
| 26 | Magistral Medium [snapshot unspecified] | 16.2% | Official leaderboard | ||
| 27 | Amazon Nova Micro | 14.5% | Dec 2024 | Official leaderboard | |
| 28 | Mistral Large [snapshot unspecified] | 13.4% | Official leaderboard |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official FinQA (Snorkel OpenEnv environment) leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official FinQA (Snorkel OpenEnv environment) leaderboard
No lab model card reports it yet
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.
More in Law, finance and medicine
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4% - HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy.
Ling-3.1-flash 81.7% - KBL In use Models answer Korean legal knowledge and reasoning tasks and Korean bar exam questions. It has 7 knowledge tasks, 4 reasoning tasks and 2,510 bar exam items.
Solar Pro 4 77.5% - KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. Solar Pro 4 93.2%