SWE MarathonBrowse 284 families
SWE Marathon
Can the model complete substantial programming work under this benchmark's agent setup?
Version / measure
- Domain
- Coding
- Ranked models
- 10
- Reported values
- 15
- Reports
- 3 · lab
- Direction
- Higher is better
- Status
- Version not specified
Rankings
Best reported score per model across effort settings
| # | Model | Best setting | Score |
|---|---|---|---|
| 1 | Kimi K3Moonshot AIOW | maxcost n/a1 value · Jul 17 · source | 42.0% |
| 2 | Claude Opus 4.8Anthropic | maxcost n/a3 values · Jul 17 · source | 40.0% |
| 3 | GPT-5.6 SolOpenAI | maxcost n/a1 value · Jul 17 · source | 39.0% |
| 4 | Claude Fable 5Anthropic | maxcost n/a2 values · Jul 17 · source | 35.0% |
| 5 | Grok 4.5SpaceXAI | reportedcost n/a1 value · Jul 16 · source | 29.0% |
| 6 | Claude Opus 4.7Anthropic | maxcost n/a1 value · Jul 16 · source | 16.0% |
| 7 | GPT-5.5OpenAI | xhighcost n/a2 values · Jul 17 · source | 14.0% |
| 8 | GLM-5.2Z.aiOW | maxcost n/a2 values · Jul 17 · source | 13.0% |
| 9 | Gemini 3.1 ProGoogle | reportedcost n/a1 value · Jun 16 · source | 4.0% |
| 10 | GLM-5.1Z.aiOW | reportedcost n/a1 value · Jun 16 · source | 1.0% |
Effort curve
Score vs. reported cost. 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 ↗