AI 精选动态
智能评分 60
GLM-5.2上线ModelScope
AI 推荐理由
提供了GLM-5.2在编码基准上的具体量化优势，尤其是开源SOTA表现和与Claude Opus 4.8的对比，值得关注其实际应用效果。核心解读
GLM-5.2在ModelScope上线，支持1M上下文，基于KV8、LayerSplit、IndexShare技术，在Code Arena全球排名第2，是FrontierSWE、SWE-Marathon、PostTrainBench上排名最高的开源模型，性能介于Claude Opus 4.7和4.8之间；编码方面Terminal-Bench 2.1开源SOTA（比GLM-5.1高17.5%），MCP-Atlas与Opus 4.8差距在0.8%内。
全文
GLM-5.2 is now live on ModelScope! Built for long-horizon tasks, with Solid 1M context and stronger coding capabilities. 🚀
🤖 https://t.co/CFom9Lzm6c
#2 on Code Arena among all available models globally. Highest-ranked open-source model on FrontierSWE, SWE-Marathon, and PostTrainBench, performance between Claude Opus 4.7 and 4.8.
🧠 Solid 1M context: KV8, LayerSplit, IndexShare
💻 Coding: Terminal-Bench 2.1 open-source SOTA (+17.5% over GLM-5.1), MCP-Atlas within 0.8% of Opus 4.8. Engineering feel between Claude Opus 4.7 and 4.8
⚡ Long-horizon execution: from a single requirement description to a fully deployable multi-platform app in one continuous run
vLLM, SGLang, Transformers supported.
