DeepSWE: leaderboard
DataCurve benchmark where frontier coding agents work through original, long-horizon software engineering tasks in realistic repositories.
Metric: Pass@1 (%). Source: deepswe.datacurve.ai. Status: years away from saturation. 31 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-6 (xHigh) | 74.1 |
| 2 | Gemini 3 Flash | 73.8 |
| 3 | Claude Opus 5 | 73.6 |
| 4 | GPT-6 | 73.2 |
| 5 | GPT-5.6 Sol | 72.7 |
| 6 | GPT-5.6 Sol (xHigh) | 70.7 |
| 7 | Claude Fable 5 | 69.9 |
| 8 | GLM-5.3 | 69 |
| 9 | Kimi K3 | 68.5 |
| 10 | Grok 4.6 | 67.5 |
| 11 | GPT-5.5 (xHigh) | 67 |
| 12 | GPT-5.5 | 64.4 |
| 13 | GLM-5.3 Flash | 63.4 |
| 14 | DeepSeek V4 Pro | 62.8 |
| 15 | Claude Opus 4.8 | 59 |
Interactive version: theaggregate.ai/benchmark?slug=deepswe · How It Works · Data refreshed daily, snapshot 2026-09-05.