SWE Atlas (All Workflows): leaderboard
Metric: Pass@1 (%): mean per-trial pass rate over all 284 SWE Atlas tasks (124 Codebase Q&A, 90 Test Writing and 70 Refactoring tasks in real repositories, scored by programmatic checks plus must-have rubrics), three trials per task run through the Harbor framework, each model in its provider's own agent harness (Codex CLI, Claude Code, Gemini CLI) or in mini-SWE-agent; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7 (xHigh) | 38.94 |
| 2 | GPT-5.4 (xHigh) | 38 |
| 3 | Claude Opus 4.6 (High) | 31.83 |
| 4 | Claude Sonnet 4.6 (High) | 31.63 |
| 5 | GLM-5 | 24.03 |
| 6 | Gemini 3.1 Pro (Preview) (High) | 23.73 |
| 7 | Kimi K2.5 | 19.05 |
| 8 | Gemini 3 Flash (High) | 15.65 |
| 9 | MiniMax-M2.5 | 15.2 |
Interactive version: theaggregate.ai/benchmark?slug=swe-atlas-all-workflows · How It Works · Data refreshed daily, snapshot 2026-10-07.