WideSWE: leaderboard
WideSWE (2026) — 120 real software-engineering tasks, 60 bug fixes and 60 features across 41 ecosystems, each requiring coordinated changes to several repositories from one shared request. A task counts as solved only when every target repository passes its hidden tests.
Metric: Task success (%) over 120 cross-repository tasks (60 bug fixes, 60 features, 41 ecosystems); a task succeeds only when every target repository passes its hidden tests (Claude Code harness). Source: zju-aces-ise.github.io. Saturation forecast: Around 2029. 6 models tracked. Updated 1 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Qwen 3.8 Max | xHigh | 37.5 | #59 |
| 2 | Claude Opus 5 | High | 35 | #20 |
| 3 | GPT-5.6 Sol | High | 32.5 | #30 |
| 4 | DeepSeek V4 Pro | High | 26.67 | #191 |
| 5 | GLM-5.3 | xHigh | 20 | #68 |
| 6 | Gemini 3.8 Flash | xHigh | 10.83 | #36 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=wideswe · How It Works · Data refreshed daily, snapshot 2026-11-16.