Can LLMs Falsify?: leaderboard
Tests whether LLMs can generate counterexamples to falsify incorrect mathematical hypotheses - a core scientific reasoning capability.
Metric: Counterexample Rate (%). Source: falsifiers.github.io. Saturation forecast: Around 2029. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 Mini (High) | 8.9 | #272 (O3 Mini) |
| 2 | DeepSeek R1 | 5.8 | #251 |
| 3 | Claude 3.5 Sonnet | 4.6 | #350 |
| 4 | DeepSeek V3 | 2.7 | #321 |
| 5 | Gemini 2.0 Flash (Thinking) | 2.1 | #342 (Gemini 2.0 Flash) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=can-llms-falsify · How It Works · Data refreshed daily, snapshot 2026-10-20.