SWE Marathon
Can the model complete substantial programming work under this benchmark's agent setup?
Coding10 ranked models15 reported values3 reportsHigher is better
Top rankings
One row per model, using its best reported score across effort settings.
| Rank | Model | Best score | Best reported setting |
|---|---|---|---|
| 1 | Kimi K3Moonshot AI | 42.0% | max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source |
| 2 | Claude Opus 4.8Anthropic | 40.0% | max effortMoonshot AI report3 values · 3 reportsJul 17, 2026 · source |
| 3 | GPT-5.6 SolOpenAI | 39.0% | max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source |
| 4 | Claude Fable 5Anthropic | 35.0% | max effortMoonshot AI report2 values · 2 reportsJul 17, 2026 · source |
| 5 | Grok 4.5SpaceXAI | 29.0% | Reported configurationSpaceXAI report1 value · 1 reportJul 16, 2026 · source |
| 6 | Claude Opus 4.7Anthropic | 16.0% | max effortSpaceXAI report1 value · 1 reportJul 16, 2026 · source |
| 7 | GPT-5.5OpenAI | 14.0% | xhigh effortMoonshot AI report2 values · 2 reportsJul 17, 2026 · source |
| 8 | GLM-5.2Z.ai | 13.0% | max effortMoonshot AI report2 values · 2 reportsJul 17, 2026 · source |
| 9 | Gemini 3.1 ProGoogle | 4.0% | Reported configurationZ.ai report1 value · 1 reportJun 16, 2026 · source |
| 10 | GLM-5.1Z.ai | 1.0% | Reported configurationZ.ai report1 value · 1 reportJun 16, 2026 · source |
Effort curve
Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.
No cost-linked effort sweep for this version.
Definition and comparison boundarypublic methodology
Can the model complete substantial programming work under this benchmark's agent setup?
Software implementation, debugging, or repository work under the published evaluation protocol. Higher is better. The value is the percentage reported in this lab's table.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
Claude Code harness.
Method / source