AA SciCode — leaderboard
Artificial Analysis independent evaluation of SciCode: 338 sub-tasks from 80 real lab problems across 16 scientific disciplines.
Metric: Accuracy (%). Source: artificialanalysis.ai. Status: saturation imminent. 538 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 58.91 |
| 2 | Kimi K3 | 58.68 |
| 3 | GPT-5.6 Sol (High) | 56.94 |
| 4 | GPT-5.4 (xHigh) | 56.6 |
| 5 | GPT-5.6 Sol (Medium) | 56.48 |
| 6 | GPT-5.5 (xHigh) | 56.13 |
| 7 | Gemini 3 Pro (Preview) (High) | 56.13 |
| 8 | GPT-5.6 Sol (Max) | 56.13 |
| 9 | GPT-5.6 Sol (xHigh) | 56.02 |
| 10 | GPT-5.5 (High) | 55.9 |
| 11 | GPT-5.6 Sol (Low) | 55.44 |
| 12 | GPT-5.2 Codex (xHigh) | 54.63 |
| 13 | Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | 54.51 |
| 14 | Grok 4.5 (High) | 54.05 |
| 15 | GPT-5.6 Terra (Max) | 53.94 |
Interactive version: theaggregate.ai/benchmark?slug=aa-scicode · How the rankings work · Data refreshed daily, snapshot 2026-07-22.