glm-5.2 hits 81.0 on terminal-bench 2.1, within a few points of claude opus 4.8, and it's open source. the 1m context that actually holds up under long agentic sessions is the part worth paying…

智谱GLM-5.2这次的核心是长程代码任务，1M无损上下文，下一步他要搞 Autonomous Agent System 完全自治的智能体 1M无损上下文，是专门针对Coding Agent长时间干活这个场景喂了大量训练数据，覆盖自动化研究、性能优化等，使得模型一次性处理88万token的完整工程，上下文不丢失不混乱 多个长程任务基准上介于Claude Opus 4.7与4.8之间…

GLM 5.2 效果确实不错, 可以比肩 GPT

Zhipu AI just released GLM 5.2 on Hugging Face A 1M-context open model for agentic engineering and advanced coding. 744B MoE with IndexShare sparse attention, cutting FLOPs by 2.9× at 1M tokens. SOTA…

A local model beat a 27B rival on one shot Pac Man. GLM 5.2 from tops that coding test: - Open weights landing in roughly a week - Same family strong on agentic benchmarks The open weights drop I am…

🐾 GLM-5.1: Next-Gen AI Coding Assistant Experience unparalleled coding assistance that gets better with time! GLM-5.1 handles complex problems like a pro and keeps improving as it works through…