Benchmark Almanac

DeepFinance

GPT-6 Astra 57.6% leads since November 2026
Not in model cards

Agents do private equity and investment banking work on 21 anonymized real deals: valuation, due diligence, underwriting and financial modeling. Each task comes with a data room of about 19 files and would take a professional about 27 hours. Rubrics of about 84 criteria grade the output, the workflow and the judgment behind each assumption.

Each lab's models run in that lab's own agent harness, such as Claude Code or Codex, with web search on, so a score reflects the model and its harness. Scores average 10 attempts per task. The top four runs are within each other's 95% intervals. The post doesn't name the grader. The tasks are not released.

Made by
Julius Research
Released
November 2026
Kind
Finance
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
Gemini 4 Carbon #2 on the capability index
Last model added
November 2026
Scores refreshed
November 12th, 2026
24
Entries on its own leaderboard, 23 models we track

GPT-6 Astra leads with 57.6%

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. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

24 entries. Scroll sideways for the rest.

  1. 57.6 GPT-6 Astra
  2. 56.6 Gemini 4 Carbon
  3. 54.8 Claude Opus 5.5
  4. 54.6 Claude Sonnet 5.5
  5. 54.3 GPT-6.1 Sol
  6. 47.2 GPT-6.1 Terra
  7. 45.6 Claude Fable 5.1
  8. 44.2 Muse Wonder 1.0
  9. 41.4 Gemini 4 Argon
  10. 40.0 Claude Haiku 5.5
  11. 37.0 Muse Spark 1.3
  12. 36.5 Muse Ember
  13. 34.6 Kimi K3.1 Pro
  14. 34.1 Grok 4.7
  15. 30.3 Qwen 3.8 Max
  16. 29.8 Inkling 2 Small
  17. 27.5 GLM-5.3
  18. 27.3 DeepSeek V4.1 Flash
  19. 26.4 GPT-6 Luna
  20. 21.5 Mistral Large 4
  21. 20.3 DeepSeek-V4-Pro
  22. 14.5 Nemotron 3 Ultra
  23. 13.1 Inkling
  24. 10.9 Inkling-Small
  1. 1 GPT-6 Astra max 57.6%
  2. 2 Gemini 4 Carbon medium 56.6%
  3. 3 Claude Opus 5.5 max 54.8%
  4. 4 Claude Sonnet 5.5 max 54.6%
  5. 5 GPT-6.1 Sol max 54.3%
  6. 6 GPT-6.1 Terra max 47.2%
  7. 7 Claude Fable 5.1 max 45.6%
  8. 8 Muse Wonder 1.0 max 44.2%
  9. 9 Gemini 4 Argon high 41.4%
  10. 10 Claude Haiku 5.5 max 40%
  11. 11 Muse Spark 1.3 max 37%
  12. 12 Muse Ember max 36.5%
  13. 13 Kimi K3.1 Pro max 34.6%
  14. 14 Grok 4.7 xhigh 34.1%
  15. 15 Qwen 3.8 Max xhigh 30.3%
  16. 16 Inkling 2 Small max 29.8%
  17. 17 GLM-5.3 max 27.5%
  18. 18 DeepSeek V4.1 Flash max 27.3%
  19. 19 GPT-6 Luna max 26.4%
  20. 20 Mistral Large 4 high 21.5%
  21. 21 DeepSeek-V4-Pro max 20.3%
  22. 22 Nemotron 3 Ultra high 14.5%
  23. 23 Inkling max 13.1%
  24. 24 Inkling-Small max 10.9%

Source: Official DeepFinance leaderboard

#EntryScoreSettingDateSource
1 GPT-6 Astra 57.6% max, Codex 0.166.0 Sep 2026 Official leaderboard
2 Gemini 4 Carbon 56.6% medium, OpenCode 1.19.3 Oct 2026 Official leaderboard
3 Claude Opus 5.5 54.8% max, Claude Code 2.1.318 Sep 2026 Official leaderboard
4 Claude Sonnet 5.5 54.6% max, Claude Code 2.1.318 Sep 2026 Official leaderboard
5 GPT-6.1 Sol 54.3% max, Codex 0.166.0 Sep 2026 Official leaderboard
6 GPT-6.1 Terra 47.2% max, Codex 0.166.0 Oct 2026 Official leaderboard
7 Claude Fable 5.1 45.6% max, Claude Code 2.1.318 Sep 2026 Official leaderboard
8 Muse Wonder 1.0 44.2% max, OpenCode 1.19.3 Nov 2026 Official leaderboard
9 Gemini 4 Argon 41.4% high, OpenCode 1.19.3 Sep 2026 Official leaderboard
10 Claude Haiku 5.5 40% max, Claude Code 2.1.318 Oct 2026 Official leaderboard
11 Muse Spark 1.3 37% max, OpenCode 1.19.3 Sep 2026 Official leaderboard
12 Muse Ember 36.5% max, OpenCode 1.19.3 Nov 2026 Official leaderboard
13 Kimi K3.1 Pro 34.6% max, Kimi Code 2.4.1 Oct 2026 Official leaderboard
14 Grok 4.7 34.1% xhigh, x.ai 1.0.70 Sep 2026 Official leaderboard
15 Qwen 3.8 Max 30.3% xhigh, Qwen Code 0.26.6 Aug 2026 Official leaderboard
16 Inkling 2 Small 29.8% max, OpenCode 1.19.3 Nov 2026 Official leaderboard
17 GLM-5.3 27.5% max, OpenCode 1.19.3 Aug 2026 Official leaderboard
18 DeepSeek V4.1 Flash 27.3% max, OpenCode 1.19.3 Sep 2026 Official leaderboard
19 GPT-6 Luna 26.4% max, Codex 0.166.0 Sep 2026 Official leaderboard
20 Mistral Large 4 21.5% high, Mistral Vibe 2.24.3 Oct 2026 Official leaderboard
21 DeepSeek-V4-Pro 20.3% max, OpenCode 1.19.3 Apr 2026 Official leaderboard
22 Nemotron 3 Ultra 14.5% high, OpenCode 1.19.3 Jun 2026 Official leaderboard
23 Inkling 13.1% max, OpenCode 1.19.3 Jul 2026 Official leaderboard
24 Inkling-Small 10.9% max, OpenCode 1.19.3 Jul 2026 Official leaderboard

Score vs. tokens

Mean rubric score against all tokens a run used per task, on a log scale. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 1M10M100M Tokens per task, log scale, lower is better Mean rubric score DeepSeek-V4-Pro (max): Mean rubric score 20.3%, tokens per task 2.6M Nemotron 3 Ultra (high): Mean rubric score 14.5%, tokens per task 2.8M Qwen 3.8 Max (xhigh): Mean rubric score 30.3%, tokens per task 12.6M GLM-5.3 (max): Mean rubric score 27.5%, tokens per task 5.6M Claude Fable 5.1 (max): Mean rubric score 45.6%, tokens per task 3.1M Muse Spark 1.3 (max): Mean rubric score 37%, tokens per task 10.9M DeepSeek V4.1 Flash (max): Mean rubric score 27.3%, tokens per task 3.9M Grok 4.7 (xhigh): Mean rubric score 34.1%, tokens per task 5.9M Claude Opus 5.5 (max): Mean rubric score 54.8%, tokens per task 10.5M GPT-6 Luna (max): Mean rubric score 26.4%, tokens per task 3.6M Claude Sonnet 5.5 (max): Mean rubric score 54.6%, tokens per task 22.6M GPT-6.1 Sol (max): Mean rubric score 54.3%, tokens per task 2.5M Gemini 4 Argon (high): Mean rubric score 41.4%, tokens per task 6.7M Mistral Large 4 (high): Mean rubric score 21.5%, tokens per task 5M Claude Haiku 5.5 (max): Mean rubric score 40%, tokens per task 16.5M Kimi K3.1 Pro (max): Mean rubric score 34.6%, tokens per task 6.1M GPT-6.1 Terra (max): Mean rubric score 47.2%, tokens per task 4.2M Gemini 4 Carbon (medium): Mean rubric score 56.6%, tokens per task 7.4M Inkling 2 Small (max): Mean rubric score 29.8%, tokens per task 6.4M Muse Ember (max): Mean rubric score 36.5%, tokens per task 7.3M Muse Wonder 1.0 (max): Mean rubric score 44.2%, tokens per task 8.2M. Not in our model list Inkling (max): Mean rubric score 13.1%, tokens per task 2.4M Inkling-Small (max): Mean rubric score 10.9%, tokens per task 1.1M GPT-6 Astra (max): Mean rubric score 57.6%, tokens per task 2.4M GPT-6 Astra (max)Inkling (max)Inkling-Small (max) 0% 20% 40% 60% 1M10M100M Tokens per task, log scale, lower is better Mean rubric score DeepSeek-V4-Pro (max): Mean rubric score 20.3%, tokens per task 2.6M Nemotron 3 Ultra (high): Mean rubric score 14.5%, tokens per task 2.8M Qwen 3.8 Max (xhigh): Mean rubric score 30.3%, tokens per task 12.6M GLM-5.3 (max): Mean rubric score 27.5%, tokens per task 5.6M Claude Fable 5.1 (max): Mean rubric score 45.6%, tokens per task 3.1M Muse Spark 1.3 (max): Mean rubric score 37%, tokens per task 10.9M DeepSeek V4.1 Flash (max): Mean rubric score 27.3%, tokens per task 3.9M Grok 4.7 (xhigh): Mean rubric score 34.1%, tokens per task 5.9M Claude Opus 5.5 (max): Mean rubric score 54.8%, tokens per task 10.5M GPT-6 Luna (max): Mean rubric score 26.4%, tokens per task 3.6M Claude Sonnet 5.5 (max): Mean rubric score 54.6%, tokens per task 22.6M GPT-6.1 Sol (max): Mean rubric score 54.3%, tokens per task 2.5M Gemini 4 Argon (high): Mean rubric score 41.4%, tokens per task 6.7M Mistral Large 4 (high): Mean rubric score 21.5%, tokens per task 5M Claude Haiku 5.5 (max): Mean rubric score 40%, tokens per task 16.5M Kimi K3.1 Pro (max): Mean rubric score 34.6%, tokens per task 6.1M GPT-6.1 Terra (max): Mean rubric score 47.2%, tokens per task 4.2M Gemini 4 Carbon (medium): Mean rubric score 56.6%, tokens per task 7.4M Inkling 2 Small (max): Mean rubric score 29.8%, tokens per task 6.4M Muse Ember (max): Mean rubric score 36.5%, tokens per task 7.3M Muse Wonder 1.0 (max): Mean rubric score 44.2%, tokens per task 8.2M. Not in our model list Inkling (max): Mean rubric score 13.1%, tokens per task 2.4M Inkling-Small (max): Mean rubric score 10.9%, tokens per task 1.1M GPT-6 Astra (max): Mean rubric score 57.6%, tokens per task 2.4M GPT-6 Astra (max)Inkling (max)Inkling-Small (max)
ModelSettingMean rubric scoreTokens per task
GPT-6 Astra max, Codex 0.166.0 57.6% 2,410,000
Gemini 4 Carbon medium, OpenCode 1.19.3 56.6% 7,420,000
Claude Opus 5.5 max, Claude Code 2.1.318 54.8% 10,500,000
Claude Sonnet 5.5 max, Claude Code 2.1.318 54.6% 22,600,000
GPT-6.1 Sol max, Codex 0.166.0 54.3% 2,520,000
GPT-6.1 Terra max, Codex 0.166.0 47.2% 4,210,000
Claude Fable 5.1 max, Claude Code 2.1.318 45.6% 3,150,000
Muse Wonder 1.0 max, OpenCode 1.19.3 44.2% 8,180,000
Gemini 4 Argon high, OpenCode 1.19.3 41.4% 6,670,000
Claude Haiku 5.5 max, Claude Code 2.1.318 40% 16,500,000
Muse Spark 1.3 max, OpenCode 1.19.3 37% 10,900,000
Muse Ember max, OpenCode 1.19.3 36.5% 7,300,000
Kimi K3.1 Pro max, Kimi Code 2.4.1 34.6% 6,080,000
Grok 4.7 xhigh, x.ai 1.0.70 34.1% 5,900,000
Qwen 3.8 Max xhigh, Qwen Code 0.26.6 30.3% 12,600,000
Inkling 2 Small max, OpenCode 1.19.3 29.8% 6,380,000
GLM-5.3 max, OpenCode 1.19.3 27.5% 5,610,000
DeepSeek V4.1 Flash max, OpenCode 1.19.3 27.3% 3,930,000
GPT-6 Luna max, Codex 0.166.0 26.4% 3,630,000
Mistral Large 4 high, Mistral Vibe 2.24.3 21.5% 5,040,000
DeepSeek-V4-Pro max, OpenCode 1.19.3 20.3% 2,610,000
Nemotron 3 Ultra high, OpenCode 1.19.3 14.5% 2,830,000
Inkling max, OpenCode 1.19.3 13.1% 2,400,000
Inkling-Small max, OpenCode 1.19.3 10.9% 1,090,000

The best score, 57.6%, costs $6.73 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.1 Sol at max comes within 3.3 points of the best for $1.17 a task, 6 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Google DeepMind Meta Cost frontier One model at each effort level
0% 20% 40% 60% $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra, max Gemini 4 Carbon, medium GPT-6.1 Terra, max Muse Ember, max GPT-6 Luna, max 0% 20% 40% 60% $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra, max GPT-6.1 Sol, max

Source: Official DeepFinance leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra max Codex 0.166.0 57.6% $6.73
Gemini 4 Carbon medium OpenCode 1.19.3 56.6% $3.12
Claude Opus 5.5 max Claude Code 2.1.318 54.8% $8.63
Claude Sonnet 5.5 max Claude Code 2.1.318 54.6% $6.94
GPT-6.1 Sol max Codex 0.166.0 54.3% $1.17
GPT-6.1 Terra max Codex 0.166.0 47.2% $0.69
Claude Fable 5.1 max Claude Code 2.1.318 45.6% $12
Gemini 4 Argon high OpenCode 1.19.3 41.4% $2.84
Claude Haiku 5.5 max Claude Code 2.1.318 40% $1.73
Muse Spark 1.3 max OpenCode 1.19.3 37% $4.87
Muse Ember max OpenCode 1.19.3 36.5% $0.13
Kimi K3.1 Pro max Kimi Code 2.4.1 34.6% $2.67
Grok 4.7 xhigh x.ai 1.0.70 34.1% $4.71
Qwen 3.8 Max xhigh Qwen Code 0.26.6 30.3% $4.99
Inkling 2 Small max OpenCode 1.19.3 29.8% $0.50
GLM-5.3 max OpenCode 1.19.3 27.5% $1.11
DeepSeek V4.1 Flash max OpenCode 1.19.3 27.3% $0.13
GPT-6 Luna max Codex 0.166.0 26.4% $0.12
Mistral Large 4 high Mistral Vibe 2.24.3 21.5% $3.69
DeepSeek-V4-Pro max OpenCode 1.19.3 20.3% $1.49
Nemotron 3 Ultra high OpenCode 1.19.3 14.5% $0.64
Inkling max OpenCode 1.19.3 13.1% $0.64
Inkling-Small max OpenCode 1.19.3 10.9% $0.15

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard 1 leader
GPT-6 Astra 57.6%

Source: Official DeepFinance leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6 Astra Claude Fable 5.1 Qwen 3.8 Max DeepSeek-V4-Pro 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6 Astra Claude Fable 5.1

Source: Official DeepFinance 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

All 179 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%