SciCode: leaderboard
SciCode scientific coding benchmark: the public leaderboard reports subproblem resolve rate, separately from main-problem completion. The full collection has 80 problems and 338 subproblems; historical source configurations and its explicitly distinguished newer Sonnet revision are preserved.
Metric: Subproblem Resolve Rate (%). Source: scicode-bench.github.io. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 Mini (High) | 34.4 | #272 (O3 Mini) |
| 2 | O3 Mini (Low) | 33.3 | #272 (O3 Mini) |
| 3 | O3 Mini (Medium) | 33 | #272 (O3 Mini) |
| 4 | DeepSeek R1 | 28.5 | #251 |
| 5 | O1 Preview | 28.5 | #193 |
| 6 | Claude 3.5 Sonnet | 26 | #350 |
| 7 | GPT-4o | 25 | #343 |
| 8 | DeepSeek V3 | 23.7 | #321 |
| 9 | GPT-4 Turbo | 22.9 | #404 |
| 10 | O1 Mini | 22.2 | #323 |
| 11 | Gemini 1.5 Pro | 21.9 | #445 |
| 12 | Claude 3 Opus | 21.5 | #571 |
| 13 | Qwen 2 72B Instruct | 17 | #576 |
| 14 | Claude 3 Sonnet | 17 | #775 |
| 15 | Mixtral 8x22B Instruct | 16.3 | #728 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scicode · How It Works · Data refreshed daily, snapshot 2026-10-20.