Overview
- Z. AI published GLM-5.2 between June 13 and June 16 and posted the full model weights on Hugging Face under an MIT license so developers can run, modify, and commercially deploy the model locally.
- The model uses a Mixture‑of‑Experts architecture that activates about 40 billion parameters per query while the full model contains roughly 744–753 billion parameters and supports a 1 million‑token context window for long, multi‑step coding tasks.
- Independent benchmark reports show GLM-5.2 narrowly outperforms GPT-5.5 on long‑horizon coding tests (about a 1% lead on FrontierSWE) and posts scores such as 81.0 on Terminal‑Bench 2.1 and 62.1 on SWE‑bench Pro, plus a 51 on the Artificial Analysis Intelligence Index for open models.
- Z. AI has priced hosted API tiers well below major incumbents and reporting indicates GLM-5.2 runs at roughly one‑sixth the cost of OpenAI’s offering, giving teams the choice to self‑host or use cheaper hosted access.
- The release tightens competition with OpenAI, Google, and Anthropic and will prompt further independent evaluations and vendor responses because MoE routing brings trade‑offs in latency, routing complexity, and evaluation variability even as it lowers per‑query compute.