Scale AI - SWE-Bench Pro: leaderboard
Extended SWE-bench testing long-horizon tasks on both public open-source and private commercial repositories. Part of Scale's SEAL evaluation platform.
Metric: Score. Source: scale.com. Status: years away from saturation. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Muse Spark 1.1 | 61.5 |
| 2 | GPT-5.4 (xHigh) | 59.1 |
| 3 | Muse Spark | 55 |
| 4 | Claude Opus 4.6 (Thinking) | 51.9 |
| 5 | Gemini 3.1 Pro (Preview) (Thinking) | 46.1 |
| 6 | Claude Opus 4.5 (20251101) | 45.89 |
| 7 | Qwen3 Coder Next | 44.3 |
| 8 | Claude Sonnet 4.5 | 43.6 |
| 9 | Gemini 3 Pro (Preview) | 43.3 |
| 10 | Claude Sonnet 4 | 42.7 |
| 11 | GPT-5 (High) | 41.78 |
| 12 | GPT-5.2 Codex | 41.04 |
| 13 | Claude Haiku 4.5 | 39.45 |
| 14 | Qwen 3 Coder 480B A35B Instruct | 38.7 |
| 15 | MiniMax-M2.1 | 36.81 |
Interactive version: theaggregate.ai/benchmark?slug=scale-ai-swe-bench-pro · How It Works · Data refreshed daily, snapshot 2026-09-05.