DeathMath - Overall: leaderboard
DeathMath (DOoM leaderboard) by Vikhr Models: complex math and physics problems posed in Russian with a Russian system prompt. The score is the mean of math and physics accuracy, 0-100, higher is better. Submission-driven; system prompts differ between rows, and a model with a failed section is left out.
Metric: Mean of math and physics accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 77.6 | #82 |
| 2 | Gemini 2.5 Pro | 72.8 | #159 |
| 3 | GPT-5 | 70.5 | #105 |
| 4 | O3 Mini (High) | 69.2 | #282 (O3 Mini) |
| 5 | GPT-OSS-120B | 67.5 | #346 |
| 6 | O3 | 66.9 | #132 |
| 7 | O4 Mini (High) | 66.4 | #185 (O4 Mini) |
| 8 | GPT-5 Nano | 64.9 | #439 |
| 9 | GPT-5 Mini | 63.4 | #193 |
| 10 | Claude Sonnet 4.5 | 61 | #148 |
| 11 | Claude Opus 4.1 | 60.7 | #146 |
| 12 | GPT-OSS-20B | 58.3 | #523 |
| 13 | Claude Sonnet 4 | 55.1 | #207 |
| 14 | QwQ-32B | 53 | #422 |
| 15 | Gemini 2.0 Flash | 51.4 | #351 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=deathmath-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.