Benchmarks / Coding
SciCode
Scientific-computing code generation from research problems.
maintainer unknown · 235 observations on record
Versions
1 version
| Version | Tasks |
|---|---|
| SciCode current scicode | — |
Current leaderboard
SciCode · Accuracy · top 25 of 235 systems
| # | System | Accuracy % |
|---|---|---|
| 1 | Gemini 3.1 Pro Preview Google DeepMind | 58.9% |
| 2 | Muse Spark 1.1xhigh effort Meta AI | 58.2% |
| 3 | GPT 5.6 Solhigh effort OpenAI | 56.9% |
| 4 | GPT 5.4 (2026-03-05)xhigh effort OpenAI | 56.6% |
| 5 | GPT 5.6 Solmedium effort OpenAI | 56.5% |
| 6 | Gemini 3 Pro Previewhigh effort Google DeepMind | 56.1% |
| 7 | GPT 5.6 Solmax effort OpenAI | 56.1% |
| 8 | GPT 5.5 (2026-04-22)xhigh effort OpenAI | 56.1% |
| 9 | GPT 5.6 Solxhigh effort OpenAI | 56.0% |
| 10 | GPT 5.5 (2026-04-22)high effort OpenAI | 55.9% |
| 11 | GPT 5.6 Sollow effort OpenAI | 55.4% |
| 12 | GPT 5.2 Codexxhigh effort OpenAI | 54.6% |
| 13 | Claude Opus 4.7max effort Anthropic | 54.5% |
| 14 | Grok 4.5high effort xAI | 54.1% |
| 15 | GPT 5.6 Terramax effort OpenAI | 53.9% |
| 16 | Claude Sonnet 5max effort Anthropic | 53.6% |
| 17 | GPT 5.5 (2026-04-22)medium effort OpenAI | 53.5% |
| 18 | Claude Opus 4.8max effort Anthropic | 53.5% |
| 19 | Kimi K2 6 Moonshot AI | 53.5% |
| 20 | GPT 5.3 Codexxhigh effort OpenAI | 53.2% |
| 21 | Gemini 3.5 Flashhigh effort Google DeepMind | 53.1% |
| 22 | Gemini 3.5 Flashmedium effort Google DeepMind | 53.0% |
| 23 | GPT 5.6 Lunamax effort OpenAI | 52.5% |
| 24 | GPT 5.2 (2025-12-11)xhigh effort OpenAI | 52.1% |
| 25 | Claude Opus 4.6max effort Anthropic | 51.9% |
One row per evaluated system — reasoning-effort variants rank separately.