DeathMath - Physics: leaderboard
DeathMath (DOoM leaderboard) by Vikhr Models, physics section: complex physics problems posed in Russian. Accuracy 0-100, higher is better. The Qwen2.5 72B Instruct run, scored 0 on physics beside 37.9 on math and 17.3 physics in the April 2025 table, is omitted as failed.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 61.2 | #82 |
| 2 | Gemini 2.5 Pro | 58.2 | #159 |
| 3 | Claude Sonnet 4.5 | 53.1 | #148 |
| 4 | Claude Opus 4.1 | 51 | #146 |
| 5 | GPT-5 | 50 | #105 |
| 6 | GPT-OSS-120B | 50 | #346 |
| 7 | O3 Mini (High) | 50 | #282 (O3 Mini) |
| 8 | Gemini 2.0 Flash | 46.9 | #351 |
| 9 | Claude Sonnet 4 | 46.9 | #207 |
| 10 | O3 | 46.9 | #132 |
| 11 | GPT-5 Nano | 45.9 | #439 |
| 12 | O4 Mini (High) | 45.9 | #185 (O4 Mini) |
| 13 | GPT-5 Mini | 41.8 | #193 |
| 14 | QwQ-32B | 40.8 | #422 |
| 15 | Claude 3.7 Sonnet | 39.8 | #259 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=deathmath-physics · How It Works · Data refreshed daily, snapshot 2026-11-05.