NVIDIA ComputeEval - Math Libs: leaderboard
NVIDIA ComputeEval zero-shot pass@1 on the 101 Math Libs problems of the 2026.1 release: cuSPARSE, cuSOLVER, cuFFT and cuRAND, graded by held-out compile-and-run tests.
Metric: Pass@1 (%, zero-shot, 2026.1 release, 101 problems). Source: huggingface.co. Saturation forecast: Around February 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (High) | 91.1 | #67 (Claude Opus 4.6) |
| 2 | Claude Sonnet 4.6 (High) | 91.1 | #94 (Claude Sonnet 4.6) |
| 3 | Claude Opus 4.6 (Medium) | 91.1 | #67 (Claude Opus 4.6) |
| 4 | Claude Sonnet 4.6 (Medium) | 89.1 | #94 (Claude Sonnet 4.6) |
| 5 | GPT-5.4 (High) | 82.2 | #83 (GPT-5.4) |
| 6 | GPT-5.4 (Medium) | 82.2 | #83 (GPT-5.4) |
| 7 | GPT-OSS-120B (High) | 57.4 | #346 (GPT-OSS-120B) |
| 8 | GPT-5 Mini (High) | 56.4 | #193 (GPT-5 Mini) |
| 9 | Kimi K2.5 (Thinking) | 55.4 | #152 (Kimi K2.5) |
| 10 | GPT-5 Mini (Medium) | 54.5 | #193 (GPT-5 Mini) |
| 11 | GPT-OSS-120B (Medium) | 48.5 | #346 (GPT-OSS-120B) |
| 12 | GLM-5 (Thinking) | 48.5 | #149 (GLM-5) |
| 13 | Gemini 3.1 Flash Lite (High) | 37.6 | #200 (Gemini 3.1 Flash Lite) |
| 14 | Gemini 3.1 Flash Lite (Medium) | 33.7 | #200 (Gemini 3.1 Flash Lite) |
| 15 | GPT-OSS-20B (Medium) | 23.8 | #523 (GPT-OSS-20B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=nvidia-computeeval-math-libs · How It Works · Data refreshed daily, snapshot 2026-11-05.