AtmosCoder-Bench: leaderboard
AtmosCoder-Bench grades quantitative atmospheric science by execution: each of 436 textbook-grounded problems is answered by a Python solve() the model writes, and the graded number is whatever that function computes, with the declared unit reconciled against ground truth. Three runs per model; accuracy (%) is the mean. Reasoning and thinking-off runs of the same checkpoint are listed separately, as upstream measures them.
Metric: Accuracy (%, mean of 3 runs). Source: github.com. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 97.6 | #31 |
| 2 | Gemini 3.1 Pro (Preview) (Thinking) | 96 | #58 (Gemini 3.1 Pro (Preview)) |
| 3 | Kimi K2.6 (Thinking) | 93.8 | #104 (Kimi K2.6) |
| 4 | DeepSeek V4 Pro (Reasoning) | 93.3 | #102 (DeepSeek V4 Pro) |
| 5 | DeepSeek V4 Flash (Reasoning) | 91.1 | #137 (DeepSeek V4 Flash) |
| 6 | GPT-5.5 (Non-reasoning) | 90.8 | #31 (GPT-5.5) |
| 7 | Kimi K2.6 (Non-reasoning) | 90.1 | #104 (Kimi K2.6) |
| 8 | Qwen 3.6 27B (Reasoning) | 88.5 | #170 (Qwen 3.6 27B) |
| 9 | DeepSeek V4 Pro (Non-reasoning) | 82.6 | #102 (DeepSeek V4 Pro) |
| 10 | DeepSeek V4 Flash (Non-reasoning) | 81.7 | #137 (DeepSeek V4 Flash) |
| 11 | Qwen 3.6 27B (Non-reasoning) | 79.4 | #170 (Qwen 3.6 27B) |
| 12 | Qwen 3.5 9B (Reasoning) | 76.9 | #379 (Qwen 3.5 9B) |
| 13 | Qwen 3.5 9B (Non-reasoning) | 63.4 | #379 (Qwen 3.5 9B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=atmoscoder-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.