Benchmarks

How Think-Reason-Learn performs on real outcomes, and the data behind FounderBrain.

VCBench

Predicting which founders succeed, scored on 4,500 founders the models never saw.

Think-Reason-LearnOthersF_0.5 · higher is better

TRL EnsembleTRL

37.9

GPT-6 SolOpenAI

33.7

Policy InductionTRL

33.0

Reasoned Rule MiningTRL

32.2

Best ML modelClassical ML

31.5

Random Rule ForestTRL

31.3

Verifiable RLTRL

29.4

Claude Sonnet 5Anthropic

27.7

GPTreeTRL

20.6

Gemini 2.5 ProGoogle

19.9

Tier-1 VCsHumans

10.7

Best entry of each kind, as of 29 September 2026. Full leaderboard on vcbench.com →

What founders and CEOs actually said, answer by answer. FounderBrain was trained on it.

Founders

13,014

answers from 266 founders in long-form interviews

CEOs

15,252

answers from 100 public-company CEOs on earnings calls