Evidence-based reasoning tasks that require weighing conflicting sources before answering.
Judgement over retrieval: what to believe when the evidence disagrees.
Newer benchmark with fewer models evaluated so far.
| # | Model | Lab | Score | Measured | Method | Source |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5 | Anthropic | 50.0% | 2026-08-07 | independent | Epoch AI Benchmarking Hub |
| 2 | Claude Fable 5 | Anthropic | 39.5% | 2026-07-17 | independent | Epoch AI Benchmarking Hub |
| 3 | GPT-5.6 Sol | OpenAI | 39.0% | 2026-08-07 | independent | Epoch AI Benchmarking Hub |
| 4 | GPT-5.5 | OpenAI | 34.3% | 2026-07-27 | independent | Epoch AI Benchmarking Hub |
| 5 | Claude Opus 4.8 | Anthropic | 28.6% | 2026-08-07 | independent | Epoch AI Benchmarking Hub |
| 6 | GPT-5.4 | OpenAI | 25.4% | 2026-06-25 | independent | Epoch AI Benchmarking Hub |
| 7 | GPT-5.2 | OpenAI | 23.0% | 2026-06-26 | independent | Epoch AI Benchmarking Hub |
| 8 | Claude Opus 4.7 | Anthropic | 19.0% | 2026-06-30 | independent | Epoch AI Benchmarking Hub |
| 9 | Claude Opus 4.5 | Anthropic | 14.3% | 2026-06-25 | independent | Epoch AI Benchmarking Hub |
| 10 | Gemini 3.1 Pro | Google DeepMind | 14.3% | 2026-06-25 | independent | Epoch AI Benchmarking Hub |
| 11 | Claude Opus 4.6 | Anthropic | 12.7% | 2026-06-29 | independent | Epoch AI Benchmarking Hub |
| 12 | GPT-5 | OpenAI | 12.7% | 2026-06-29 | independent | Epoch AI Benchmarking Hub |
| 13 | GLM-5.2open | Z.ai | 9.5% | 2026-06-29 | independent | Epoch AI Benchmarking Hub |
| 14 | Qwen3.7-Max | Alibaba | 9.5% | 2026-06-25 | independent | Epoch AI Benchmarking Hub |
| 15 | Claude Opus 4.1 | Anthropic | 7.9% | 2026-06-25 | independent | Epoch AI Benchmarking Hub |
| 16 | Gemini 3.5 Flash | Google DeepMind | 4.8% | 2026-06-25 | independent | Epoch AI Benchmarking Hub |
| 17 | Claude Sonnet 4.5 | Anthropic | 2.4% | 2026-06-25 | independent | Epoch AI Benchmarking Hub |
| 18 | Kimi K2.6open | Moonshot AI | 2.4% | 2026-06-25 | independent | Epoch AI Benchmarking Hub |