GLM-5.2: Built for Long-Horizon Tasks
Original title:GLM-5.2: Built for Long-Horizon Tasks
AI summary
GLM-5.2 has been released, supporting a 1M token context and adopting the IndexShare architecture, in which every 4 sparse attention layers share a lightweight indexer, reducing FLOPs per token at 1M context by 2.9x; improvements to the MTP layer increase speculative decoding acceptance length by 20%. On long-horizon coding benchmarks, FrontierSWE trails Opus 4.8 by only 1% and leads GPT-5.5 by 1%; PostTrainBench ranks second only to Opus 4.8; SWE-Marathon trails Opus 4.8 by 13% but ranks second. On the standard coding test Terminal-Bench 2.1 it scores 81.0 (GLM-5.1 scored 63.5), close to Opus 4.8's 85.0. The model introduces effort-level control to balance performance and latency. It is released under the MIT open-source license with no regional restrictions.
Key Takeaways
- 2 has been released,supporting a 1M token context and adopting the IndexShare architecture,in…
- improvements to the MTP layer increase speculative decoding acceptance length by 20%.
- On long-horizon coding benchmarks,FrontierSWE trails Opus 4.
- 8 by only 1% and leads GPT-5.
- PostTrainBench ranks second only to Opus 4.
Quick Facts
- Published
- Updated
- Source
- Hugging Face 社区博客(混合发现)
- Category
- Model Releases
- AI score
- 83
GLM-5.2 是第一个在长程编码基准上直逼 Opus 4.8 的开源模型,MIT 许可无地域限制,开源生态终于有了能打硬仗的长上下文工具。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: GLM-5.2: Built for Long-Horizon Tasks. June 17, 2026. https://new.nat6.net/en/news/cmqhvpfla0354slf0g5ldyctw/