Benchmark Almanac

VCBench

Think-Reason-Learn Ensemble 37.9% tops its leaderboard GPT-6 Sol 33.7% leads the models we track
Not in model cards

Models predict venture outcomes from founder profiles. The benchmark emphasizes precision through the F0.5 score. The hub adaptation substitutes synthetic profiles for the original data.

In May 2026, GPT-4o led the original benchmark with an F0.5 score of 25.1%. DeepSeek V3 had the highest precision at 59.1%.

Released
September 2025
Kind
Finance
Level
Expert
Scored
Checked automatically
Answers
Prediction
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Law and finance
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
31
Entries on its own leaderboard, 10 models we track

Think-Reason-Learn Ensemble tops the leaderboard with 37.9% GPT-6 Sol leads with 33.7% among models released by September 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 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. The best model we track is GPT-6 Sol, at 33.7%. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Think-Reason-Learn Ensemble 37.9%
  2. 2 IdeaPulley Arlo 2 35.1%
  3. 3 Investinor Linear v1 33.8%
  4. 4 GPT-6 Sol 33.7%
  5. 5 Policy Induction 33%
  6. 6 GemVC-v0 32.9%
  7. 7 Reasoned Rule Mining 32.2%
  8. 8 Pvalyou Founder Model (ML) 32.1%
  9. 9 IdeaPulley Arlo 1 31.3%
  10. 10 Random Rule Forest 31.3%
  11. 11 Verifiable-RL 29.4%
  12. 12 Structured-Rule-Stump 28.1%
  13. 13 verifiable-reasoning 27.7%
  14. 14 Claude Sonnet 5 27.7%
  15. 15 large-founder-model-v0 27.2%
  16. 16 GPT-6 Luna 26.1%
  17. 17 GPT-4o (snapshot unspecified) 25.7%
  18. 18 FinGPT-VC2 24.9%
  19. 19 Claude Opus 5.5 24.7%
  20. 20 GPT-6 Astra 24.3%
  21. 21 GPT-4o-mini (snapshot unspecified) 23%
  22. 22 Claude Haiku 4.5 22.6%
  23. 23 FinGPT-VC1 22.2%
  24. 24 o3 21.5%
  25. 25 GPTree 20.6%
  26. 26 Gemini 2.5 Pro (snapshot unspecified) 19.9%
  27. 27 DeepSeek-Reasoner 18.4%
  28. 28 Claude 3.5 Haiku 18.2%
  29. 29 GPT-5 16.2%
  30. 30 Gemini 2.5 Flash Sep 2025 14.9%
  31. 31 DeepSeek-Chat 12.1%

Source: Official VCBench leaderboard

#EntryScoreSettingSubmitted byDateLinkSource
1 Think-Reason-Learn Ensemble 37.9% Vela + Oxford vcbench.com Official leaderboard
2 IdeaPulley Arlo 2 35.1% IdeaPulley vcbench.com Official leaderboard
3 Investinor Linear v1 33.8% Investinor vcbench.com Official leaderboard
4 GPT-6 Sol 33.7% OpenAI Sep 2026 vcbench.com Official leaderboard
5 Policy Induction 33% Vela + Oxford vcbench.com Official leaderboard
6 GemVC-v0 32.9% Independent vcbench.com Official leaderboard
7 Reasoned Rule Mining 32.2% Vela + Oxford vcbench.com Official leaderboard
8 Pvalyou Founder Model (ML) 32.1% Pvalyou vcbench.com Official leaderboard
9 IdeaPulley Arlo 1 31.3% IdeaPulley vcbench.com Official leaderboard
10 Random Rule Forest 31.3% Vela + Oxford vcbench.com Official leaderboard
11 Verifiable-RL 29.4% Vela + Oxford vcbench.com Official leaderboard
12 Structured-Rule-Stump 28.1% Independent vcbench.com Official leaderboard
13 verifiable-reasoning 27.7% Vela + Oxford vcbench.com Official leaderboard
14 Claude Sonnet 5 27.7% Anthropic Jun 2026 vcbench.com Official leaderboard
15 large-founder-model-v0 27.2% Vela + Oxford vcbench.com Official leaderboard
16 GPT-6 Luna 26.1% OpenAI Sep 2026 vcbench.com Official leaderboard
17 GPT-4o (snapshot unspecified) 25.7% OpenAI vcbench.com Official leaderboard
18 FinGPT-VC2 24.9% Columbia vcbench.com Official leaderboard
19 Claude Opus 5.5 24.7% Anthropic Sep 2026 vcbench.com Official leaderboard
20 GPT-6 Astra 24.3% OpenAI Sep 2026 vcbench.com Official leaderboard
21 GPT-4o-mini (snapshot unspecified) 23% OpenAI vcbench.com Official leaderboard
22 Claude Haiku 4.5 22.6% Anthropic Oct 2025 vcbench.com Official leaderboard
23 FinGPT-VC1 22.2% Columbia vcbench.com Official leaderboard
24 o3 21.5% OpenAI Apr 2025 vcbench.com Official leaderboard
25 GPTree 20.6% Vela + Oxford vcbench.com Official leaderboard
26 Gemini 2.5 Pro (snapshot unspecified) 19.9% Google vcbench.com Official leaderboard
27 DeepSeek-Reasoner 18.4% DeepSeek vcbench.com Official leaderboard
28 Claude 3.5 Haiku 18.2% Anthropic Oct 2024 vcbench.com Official leaderboard
29 GPT-5 16.2% OpenAI Aug 2025 vcbench.com Official leaderboard
30 Gemini 2.5 Flash Sep 2025 14.9% Google Sep 2025 vcbench.com Official leaderboard
31 DeepSeek-Chat 12.1% DeepSeek vcbench.com Official leaderboard
#ModelScoreSettingSubmitted byDateLinkSource
1 GPT-6 Sol 33.7% OpenAI Sep 2026 vcbench.com Official leaderboard
2 Claude Sonnet 5 27.7% Anthropic Jun 2026 vcbench.com Official leaderboard
3 GPT-6 Luna 26.1% OpenAI Sep 2026 vcbench.com Official leaderboard
4 Claude Opus 5.5 24.7% Anthropic Sep 2026 vcbench.com Official leaderboard
5 GPT-6 Astra 24.3% OpenAI Sep 2026 vcbench.com Official leaderboard
6 Claude Haiku 4.5 22.6% Anthropic Oct 2025 vcbench.com Official leaderboard
7 o3 21.5% OpenAI Apr 2025 vcbench.com Official leaderboard
8 Claude 3.5 Haiku 18.2% Anthropic Oct 2024 vcbench.com Official leaderboard
9 GPT-5 16.2% OpenAI Aug 2025 vcbench.com Official leaderboard
10 Gemini 2.5 Flash Sep 2025 14.9% Google Sep 2025 vcbench.com Official leaderboard

Accuracy versus cost

The chart uses the cost per 1,000 founders shown by VCBench, including its estimates. 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% 10% 20% 30% 40% $0$2.5$5$7.5$10$12.5 Cost per 1,000 founders ($), lower is better F0.5 Think-Reason-Learn EnsembleIdeaPulley Arlo 2Investinor Linear v1 0% 10% 20% 30% 40% $0$2.5$5$7.5$10$12.5 Cost per 1,000 founders ($), lower is better F0.5 Think-Reason-Learn EnsembleIdeaPulley Arlo 2Investinor Linear v1
ModelSettingF0.5Cost per 1,000 founders
Think-Reason-Learn Ensemble 37.9% $0.87
IdeaPulley Arlo 2 35.1% $0.67
Investinor Linear v1 33.8% $0
GPT-6 Sol 33.7% $1.45
Policy Induction 33% $0.3
GemVC-v0 32.9% $2.87
Reasoned Rule Mining 32.2% $0.31
Pvalyou Founder Model (ML) 32.1% $0
IdeaPulley Arlo 1 31.3% $0.1
Random Rule Forest 31.3% $0.13
Verifiable-RL 29.4% $0
Structured-Rule-Stump 28.1% $0
verifiable-reasoning 27.7% $0
Claude Sonnet 5 27.7% $2.2
large-founder-model-v0 27.2% $0
GPT-6 Luna 26.1% $0.11
GPT-4o (snapshot unspecified) 25.7% $1.55
Claude Opus 5.5 24.7% $5.33
GPT-6 Astra 24.3% $7.49
GPT-4o-mini (snapshot unspecified) 23% $0.09
Claude Haiku 4.5 22.6% $0.97
o3 21.5% $9.25
GPTree 20.6% $0.12
Gemini 2.5 Pro (snapshot unspecified) 19.9% $11.51
DeepSeek-Reasoner 18.4% $1.36
Claude 3.5 Haiku 18.2% $0.52
GPT-5 16.2% $11.23
Gemini 2.5 Flash Sep 2025 14.9% $2.87
DeepSeek-Chat 12.1% $0.17

Accuracy versus cost

The chart uses the cost per 1,000 founders shown by VCBench, including its estimates. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

20% 30% 40% $0$2.5$5$7.5$10$12.5 Cost per 1,000 founders ($), lower is better F0.5 GPT-6 SolGPT-6 Luna 20% 30% 40% $0$2.5$5$7.5$10$12.5 Cost per 1,000 founders ($), lower is better F0.5 GPT-6 SolGPT-6 Luna
ModelSettingF0.5Cost per 1,000 founders
GPT-6 Sol 33.7% $1.45
Claude Sonnet 5 27.7% $2.2
GPT-6 Luna 26.1% $0.11
Claude Opus 5.5 24.7% $5.33
GPT-6 Astra 24.3% $7.49
Claude Haiku 4.5 22.6% $0.97
o3 21.5% $9.25
Claude 3.5 Haiku 18.2% $0.52
GPT-5 16.2% $11.23
Gemini 2.5 Flash Sep 2025 14.9% $2.87

The best score, 33.7%, costs $0.001 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Cost frontier One model at each effort level
0% 10% 20% 30% 40% $0.0001 $0.001 $0.01 $0.10 Cost per task, log scale GPT-6 Sol GPT-6 Luna 0% 10% 20% 30% 40% $0.0001 $0.001 $0.01 $0.10 Cost per task, log scale GPT-6 Sol GPT-6 Luna

Source: Official VCBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Sol 33.7% $0.001 372
Claude Sonnet 5 27.7% $0.002 580
GPT-6 Luna 26.1% $0.0001 444
Claude Opus 5.5 24.7% $0.005 628
GPT-6 Astra 24.3% $0.007 377
Claude Haiku 4.5 22.6% $0.001 445
o3 21.5% $0.009 1,369
Claude 3.5 Haiku 18.2% $0.0005 358
GPT-5 16.2% $0.01 1,371
Gemini 2.5 Flash Sep 2025 14.9% $0.003 1,397

The best score, 33.7%, costs $0.001 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Cost frontier One model at each effort level
0% 10% 20% 30% 40% $0.0001 $0.001 $0.01 $0.10 Cost per task, log scale GPT-6 Sol GPT-6 Luna 0% 10% 20% 30% 40% $0.0001 $0.001 $0.01 $0.10 Cost per task, log scale GPT-6 Sol GPT-6 Luna

Source: Official VCBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Sol 33.7% $0.001 372
Claude Sonnet 5 27.7% $0.002 580
GPT-6 Luna 26.1% $0.0001 444
Claude Opus 5.5 24.7% $0.005 628
GPT-6 Astra 24.3% $0.007 377
Claude Haiku 4.5 22.6% $0.001 445
o3 21.5% $0.009 1,369
Claude 3.5 Haiku 18.2% $0.0005 358
GPT-5 16.2% $0.01 1,371
Gemini 2.5 Flash Sep 2025 14.9% $0.003 1,397

4 models have held number one on the official leaderboard since its release in September 2025

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

OpenAI Anthropic
Own leaderboard Not tested since Sep 2026
GPT-6 Sol 33.7%

Source: Official VCBench leaderboard

Scores over time

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

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-6 Sol Claude Sonnet 5 Claude Haiku 4.5 o3 Claude 3.5 Haiku 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-6 Sol Claude Haiku 4.5

Source: Official VCBench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Law, finance and medicine

All 637 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%
  • 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. o1-preview 92.7%
  • Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (31 entries).
  • Pass Models scored 10 models we track, plus 21 other entries on its leaderboard.
  • Pass Newest score The newest is from September 2026, for Claude Opus 5.5.
  • Note Frontier models Only 2 of the top 10 have a score. The strongest tested is GPT-6 Astra (#4).
  • Pass Headroom The best is 33.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on vcbench.com: 10 rows for models we track and 21 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check