SWE-Marathon — Distinct named long-horizon SWE benchmark, GLM-5.2 card
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | GLM-5.3 | Z.ai | 42.5 | 1 | 2026-08-23 |
| 2 | Kimi K3 | Moonshot | 42 | 2 | 2026-08-23 |
| 3 | Claude Opus 4.8 | Anthropic | 40 | 3 | 2026-07-27 |
| 4 | GPT-5.6 Sol | OpenAI | 39 | 1 | 2026-07-27 |
| 5 | Claude Fable 5 | Anthropic | 35 | 1 | 2026-07-27 |
| 6 | Grok4.5 | SpaceXAI | 29 | 1 | 2026-08-23 |
| 7 | GPT-5.5 | OpenAI | 14 | 3 | 2026-07-27 |
| 8 | GLM-5.2 Full Open Source | Z.ai | 13 | 6 | 2026-08-23 |
| 9 | Gemini 3.1 Pro | 4 | 2 | 2026-07-13 | |
| 10 | GLM-5.1 | Z.ai | 1 | 2 | 2026-07-13 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.