GLM-5.2: Open Weights, Near-Frontier Intelligence — Zixuan Li, Z.ai

Zixuan Li, head of Z.ai, introduces GLM-5.2 to the world. On the hardest long-horizon coding and agentic benchmarks, it lands between Claude Opus 4.7 and 4.8.

Even its non-thinking mode beats GLM-5.1 with thinking turned on, and it leads other open-weight models on the Artificial Analysis Intelligence Index. Li also explains where the GLM name comes from (a 2021 paper that put Z.ai among the earliest LLM labs), and why GLM is more than a coding model. The core of his talk is why Z.ai releases near-frontier models as open weights. It lets enterprises and governments run models on their own servers, lets companies fine-tune for law, finance and security, and lets customers co-design where the models go next. He closes with one more thing: Z Code, Z.ai's own coding harness built for GLM-5.2 that also works with other frontier models. Related links: Z.ai: https://z.ai Timestamps: 0:00 Intro from swyx 0:47 Z.ai and GLM 2:05 Where the GLM name comes from 3:10 Intelligence beyond IQ 4:25 GLM-5.2 benchmarks vs.

Mer att läsa

Building advertising for the way people use AI

OpenAI · för 2 tim sedan

Server-Side Code Execution Tools for AI Agents, Compared

OpenRouter · för 12 tim sedan

v0.40.0

Ollama · för 12 tim sedan

Google froze its open source bug bounty program due to a ‘significant rise’ in AI submissions

TechCrunch AI · för 16 tim sedan