SWE Atlas (All Workflows) - Pass^3: leaderboard
Metric: Pass^3 (%): share of tasks where all three trials pass (consistency), 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 July 2028. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (xHigh) | 28.9 |
| 2 | Claude Opus 4.7 (xHigh) | 27.1 |
| 3 | GPT-5.3 Codex (xHigh) | 24.3 |
| 4 | Claude Opus 4.6 (High) | 21.1 |
| 5 | Claude Sonnet 4.6 (High) | 14.4 |
| 6 | GLM-5 | 11.6 |
| 7 | Gemini 3.1 Pro (Preview) (High) | 11.3 |
| 8 | Kimi K2.5 | 9.9 |
| 9 | Gemini 3 Flash (High) | 7.7 |
| 10 | MiniMax-M2.5 | 6 |
Interactive version: theaggregate.ai/benchmark?slug=swe-atlas-all-workflows-pass-pow-3 · How It Works · Data refreshed daily, snapshot 2026-10-07.