ProofBench
Each model’s result as Epoch AI’s Benchmarking Hub publishes it, run by Epoch or collected from the source it cites. Facthory has not run them.
Data: Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks, licensed CC BY 4.0. Read on 16 Nov 2026.
The chart shows the 25 best of 85 results; the table lists them all.
| # | Model | Organization | Result | Evidence |
|---|---|---|---|---|
| 1 | Claude Fable 5.1 (max) API access | Anthropic | 100% | – |
| 2 | Claude Fable 5.2 (max) API access | Anthropic | 100% | – |
| 3 | Claude Opus 5.5 (max) API access | Anthropic | 100% | – |
| 4 | Claude Sonnet 5.5 (max) API access | Anthropic | 100% | – |
| 5 | Claude Opus 5 (max) API access | Anthropic | 99% | – |
| 6 | Gemini 4 Argon (unknown) API access | Google DeepMind | 99% | – |
| 7 | GPT-6 Astra (unknown thinking) API access | OpenAI | 99% | – |
| 8 | GPT-6.1 Sol (unknown thinking) API access | OpenAI | 99% | – |
| 9 | Claude Fable 5 (max) API access | Anthropic | 95% | – |
| 10 | Kimi K3.1 Pro (unknown) Open weights (restricted use) | Moonshot | 91% | – |
| 11 | Claude Haiku 5.5 (max) API access | Anthropic | 89% | – |
| 12 | Kimi K3 (unknown) Open weights (non-commercial) | Moonshot |
