Benchmark scores get restated when harnesses change or models are re-run. This page keeps the trail instead of quietly overwriting it.
No historical snapshots recorded yet — the current scores are shown below.
| # | Model | Score | Observed |
|---|---|---|---|
| 1 | Claude Opus 5 | 37% | 2026-10-19 |
| 2 | DeepSeek V4.1-Flash 1017 | 29.1% | 2026-10-20 |
| 3 | GPT 5.6-Sol | 23% | 2026-10-19 |
| 4 | DeepSeek V4.1-Flash 0910 | 20.3% | 2026-10-19 |
| 5 | GLM 5.3 | 19% | 2026-10-19 |
| 6 | Kimi K3 | 17.5% | 2026-10-19 |
| 7 | DeepSeek V4-Pro 0813 | 15.5% | 2026-10-19 |
Nothing has been overwritten yet.