CRED v0.92
Science · 2026-07-20
CRED v0.92 evaluates automated verifiers on locating known errors injected into 200 copies of 100 self-reproducing empirical policy papers. Each configuration reviews the manuscript, rendered tables, analysis code, bibliography, and reproduction diagnosis; deterministic string matching credits findings that quote the changed content. The headline score is pooled single-error detection recall, with failed calls and refusals counted as misses.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 99.0 |
| Kimi K3 | 98.0 |
| GPT-5.6 Terra | 97.0 |
| GPT-5.5 | 96.5 |
| Claude Fable 5 | 96.5 |
| GPT-5.6 Luna | 95.0 |
| GLM-5.2 | 89.5 |
| Kimi K2.6 | 85.5 |
| Opus 4.8 | 85.0 |
| Sonnet 4.6 | 83.5 |
| DeepSeek-V4-Pro | 83.5 |
| DeepSeek-V4-Flash | 80.5 |
| Opus 4.1 | 78.0 |
| Sonnet 5 | 77.5 |
| Sonnet 4.5 | 77.0 |
| Opus 4 | 76.5 |
| Opus 4.5 | 73.0 |
| GLM-5 | 71.5 |
| Nemotron 3 Ultra 550B A55B | 71.5 |
| MiniMax M3 | 71.0 |
| Kimi K2.5 | 69.0 |
| Gemini 3.5 Flash | 68.5 |
| Ring 2.6 1T | 67.5 |
| Kimi K2 Thinking | 65.0 |
| GLM-4.7 | 61.0 |
| Kimi K2 Instruct | 58.0 |
| Qwen3.5 397B A17B | 55.0 |
| DeepSeek-R1-0528 | 53.0 |
| Hy3 preview | 50.5 |
| GLM 4.6 | 49.0 |
| Nemotron 3 Super 120B A12B | 46.5 |
| GLM-4.5 | 45.5 |
| gpt-oss-120b | 45.0 |
| DeepSeek-V3.2 | 41.0 |
| DeepSeek-R1 | 39.0 |
| Gemma 4 31B IT | 36.5 |
| DeepSeek-V3.1 | 34.5 |
| Mistral Large 3 675B Instruct 2512 | 32.0 |
| MiniMax M2 | 31.0 |
| O1 | 30.0 |