DeepSWE-mini (16 Tasks): leaderboard
The 16 DeepSWE v1.1 tasks selected to reproduce the full 113-task ranking at about a seventh of the cost (Spearman 0.944 on held-out configurations). A model's score here is its own number on a different, much smaller task set, not its DeepSWE score. Pass@1, higher is better. Source: huggingface.co/datasets/LocalLLaMA/deepswe-mini.
Metric: Pass@1 (%, 16-task subset). Source: huggingface.co. Saturation forecast: Around February 2027. 70 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (xHigh) | 81.25 | #6 (GPT-6 Astra) |
| 2 | GPT-6 Astra (Max) | 79.69 | #6 (GPT-6 Astra) |
| 3 | GPT-6 Astra (High) | 79.69 | #6 (GPT-6 Astra) |
| 4 | GPT-6 Astra (Medium) | 79.69 | #6 (GPT-6 Astra) |
| 5 | GPT-5.6 Sol (Max) | 78.12 | #22 (GPT-5.6 Sol) |
| 6 | Gemini 3.8 Flash (High) | 75 | #30 (Gemini 3.8 Flash) |
| 7 | GPT-5.6 Sol (High) | 75 | #22 (GPT-5.6 Sol) |
| 8 | Claude Opus 5 (Medium) | 75 | #16 (Claude Opus 5) |
| 9 | Claude Opus 5 (High) | 72.92 | #16 (Claude Opus 5) |
| 10 | Claude Opus 5 (xHigh) | 72.92 | #16 (Claude Opus 5) |
| 11 | Claude Opus 5 (Max) | 71.88 | #16 (Claude Opus 5) |
| 12 | Gemini 3.8 Flash (Medium) | 71.88 | #30 (Gemini 3.8 Flash) |
| 13 | Kimi K3 (Max) | 70.31 | #35 (Kimi K3) |
| 14 | GPT-5.5 (xHigh) | 70.31 | #33 (GPT-5.5) |
| 15 | GLM-5.3 (Max) | 70.31 | #44 (GLM-5.3) |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=deepswe-mini-16-tasks · How It Works · Data refreshed daily, snapshot 2026-11-05.