SEAL - SWE-Bench Pro (Public Dataset): leaderboard
731 real-world software engineering tasks from GPL-licensed open-source repos, testing AI agents on long-horizon bug fixing and feature implementation.
Metric: Score. Source: scale.com. Source: huggingface.co. Saturation forecast: Around January 2027. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Spark 1.1 | 61.5 | #52 |
| 2 | GPT-5.4 (xHigh) | 59.1 | #79 (GPT-5.4) |
| 3 | Muse Spark | 55 | #64 |
| 4 | Claude Opus 4.6 (Thinking) | 51.9 | #63 (Claude Opus 4.6) |
| 5 | Gemini 3.1 Pro (Preview) (Thinking) | 46.1 | #58 (Gemini 3.1 Pro (Preview)) |
| 6 | Claude Opus 4.5 (20251101) | 45.89 | #74 |
| 7 | Qwen3 Coder Next | 44.3 | #326 |
| 8 | Claude Sonnet 4.5 | 43.6 | #142 |
| 9 | Gemini 3 Pro (Preview) | 43.3 | #66 |
| 10 | Claude Sonnet 4 | 42.7 | #199 |
| 11 | GPT-5 (2025-08-07) (High) | 41.78 | |
| 12 | GPT-5.2 Codex | 41.04 | #95 |
| 13 | Claude Haiku 4.5 | 39.45 | #280 |
| 14 | Qwen 3 Coder 480B A35B Instruct | 38.7 | #315 |
| 15 | Qwen 3 Coder 480B A35B | 38.7 | #279 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seal-swe-bench-pro-public-dataset · How It Works · Data refreshed daily, snapshot 2026-10-20.