DeepSWE — Agentic SWE task-completion benchmark from GLM-5.2 cards
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Sol | OpenAI | 73 | 2 | 2026-08-23 |
| 2 | Claude Fable 5 | Anthropic | 70 | 1 | 2026-07-27 |
| 3 | GPT-5.5 | OpenAI | 70 | 3 | 2026-07-27 |
| 4 | GPT-5.6 Terra | OpenAI | 69.6 | 1 | 2026-08-23 |
| 5 | Kimi K3 | Moonshot | 67.5 | 4 | 2026-08-23 |
| 6 | GPT-5.6 Luna | OpenAI | 67.2 | 1 | 2026-08-23 |
| 7 | DeepSeek-V4-Flash-Vision-Exp | DeepSeek | 59.3 | 1 | 2026-08-23 |
| 8 | Claude Opus 4.8 | Anthropic | 59 | 5 | 2026-08-13 |
| 9 | DeepSeek-V4-Flash | DeepSeek | 54.4 | 6 | 2026-08-23 |
| 10 | Grok4.5 | SpaceXAI | 53 | 1 | 2026-08-23 |
| 11 | Gemini 3.6 Flash | 49 | 3 | 2026-08-04 | |
| 12 | GLM-5.2 Full Open Source | Z.ai | 46.2 | 7 | 2026-08-23 |
| 13 | laguna-s-2.1 | poolside | 40.4 | 1 | 2026-07-23 |
| 14 | Gemini 3.5 Flash Cyber | 37 | 1 | 2026-08-04 | |
| 15 | Gemini 3.5 Flash | 37 | 1 | 2026-08-04 | |
| 16 | Gemini 3.5 Flash Lite | 37 | 1 | 2026-08-04 | |
| 17 | Hy3-preview | Tencent | 28 | 1 | 2026-08-23 |
| 18 | MiniMax M3 | MiniMax | 20 | 2 | 2026-07-13 |
| 19 | GLM-5.1 | Z.ai | 18 | 2 | 2026-07-13 |
| 20 | Qwen3.7 Max | Alibaba | 18 | 2 | 2026-07-17 |
| 21 | DeepSeek-V4-Pro | DeepSeek | 12.8 | 4 | 2026-08-13 |
| 22 | Gemini 3.1 Pro | 10 | 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.