SWE-rebench adds five-language coding benchmark, GLM-5.2 tops open weights
Fabulous_Pollution10 · reddit · 2026-07-29
SWE-rebench released a major multilingual update for real-world software engineering tasks across five languages: Go, Java, Python, Rust, and TypeScript.
- The new leaderboard adds a multilingual slice beyond the original Python focus.
- Among open-weight models, GLM-5.2 [high] leads with 62.9% Pass@1 and 81.1% Pass@5.
- Other results shown include MiniMax M3, MiMo V2.5 Pro, DeepSeek-V4 Pro, Qwen3.6-27B, Qwen3.6-35B-A3B, and Qwen3.5-35B-A3B.
- The team says the next update in 3–4 weeks will focus heavily on models suitable for local deployment.
- They are asking for suggestions from people actually using local models for software development or coding agents, and the dataset is available for running your own agents.
More from Research
- Digital-first GLP-1 program delivers 16.1 kg average weight loss across 8,000+ women — EricTopol · 2026-10-16
- Microsoft AI explains how small-model ladders guided its 2.8T-parameter MAI-Base-2 — hboelman · 2026-10-16
- Injecting hidden-state memory cartridges into frozen Mistral-7B and Qwen2.5-7B, weights untouched — Nearby_Indication474 · 2026-10-16
- Golden Gate Claude explainer resurfaces: interpretability premise still holds — HarperSCarroll · 2026-10-16
- DeepMind: four Gemini agents beat Minecraft end-to-end in 27h14m with no human help — omarsar0 · 2026-10-16
- Is AI Interpretability Close to a Breakthrough? A New Essay Has Researchers Divided — The Decoder · 2026-10-16