Z.AI GLM-5.2: 753B open-weights model with 1M context, beats GPT-5.5 on coding at 1/6 cost
Key Questions
What is Z.AI's GLM-5.2 model?
GLM-5.2 is a 753B parameter open-weights model released by Z.AI (formerly Zhipu) under an MIT license, featuring a 1M token context window and IndexShare architecture.
How does GLM-5.2 perform compared to GPT-5.5?
It outperforms GPT-5.5 on SWE-bench and FrontierSWE coding benchmarks while costing one-sixth as much, with strong reported gains in coding and agentic tasks.
Where is the GLM-5.2 model available?
The model weights, including an FP8 variant, are available on Hugging Face under the MIT license.
What reasoning features does GLM-5.2 include?
It supports two levels of reasoning effort and is designed for long-context and coding applications.
Are independent benchmarks available for GLM-5.2?
No independent benchmarks have been released yet, though hands-on coding tests and first-look videos confirm its claimed advantages over GPT-5.5.
Z.AI (ex-Zhipu) released GLM-5.2, a 753B parameter open-weights model under MIT license with 1M token context window. It outperforms GPT-5.5 on SWE-bench and FrontierSWE at one-sixth the cost. Features IndexShare architecture and two reasoning effort levels. Available on Hugging Face with FP8 variant. No independent benchmarks yet, but strong coding/agentic gains reported. This is a major open-source contender in the coding domain. A video first look provides practical demos and reinforces the model's significance. Today a hands-on coding test video directly validates the model's claims against GPT-5.5, confirming its SWE-bench and cost advantages.