New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
SciCode benchmark
AI model leaderboard for the SciCode benchmark. Compare how large language models score on SciCode, see the full ranking, and understand what this AI benchmark measures. Grok 4 currently leads with 45.7%. Scientific code generation benchmark across physics, biology, chemistry, mathematics.
Leaderboard
| # | Model | Organization | Score | Variant | Source |
|---|---|---|---|---|---|
| #1 | Grok 4 | xAI | 45.7% | official ↗ | |
| #2 | Kimi K2 Thinking | Moonshot | 44.8% | thinking-no-tools | official ↗ |
| #3 | NVIDIA Nemotron 3 Ultra | NVIDIA | 44.6% | official ↗ | |
| #4 | Qwen3-235B-A22B-Thinking (Jul 2025) | Qwen | 42.9% | official ↗ | |
| #5 | gemini-2.5-pro | Google DeepMind | 42.8% | official ↗ | |
| #6 | GLM 4.5 | Zhipu AI | 41.7% | official ↗ | |
| #7 | o3 | OpenAI | 41% | official ↗ | |
| #8 | DeepSeek-R1-0528 | DeepSeek | 40.3% | official ↗ | |
| #9 | Claude Opus 4 | Anthropic | 39.8% | official ↗ | |
| #10 | K2 Think | Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) | 39.2% | sub-problems | official ↗ |
| #11 | gpt-oss-20b | OpenAI | 37.9% | sub-problems | official ↗ |
| #12 | GLM-4.5-Air | Zhipu AI | 37.3% | official ↗ | |
| #13 | Kimi K2 | Moonshot | 30.7% | k2-0905-no-tools-cited-k2-thinking | official ↗ |
| #14 | Qwen3-30B-A3B | Qwen | 28.5% | sub-problems | official ↗ |
Frequently asked questions about SciCode
What is the SciCode benchmark?
Scientific code generation benchmark across physics, biology, chemistry, mathematics.
How is the SciCode benchmark scored?
SciCode is scored using the pass-rate (%) metric, where a higher score is better. The maximum achievable score is 100.000. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.
Which AI model scores highest on SciCode?
As of the latest reported scores on GenAIList, Grok 4 achieves the highest result on SciCode with a score of 45.7%.
Is a higher SciCode score better?
Yes. On SciCode a higher score indicates better performance, so models near the top of the leaderboard are the strongest.