GLM-5.2 is Z.ai's open-weights reasoning model, released in June 2026, aimed at developers and enterprises that need long-horizon coding, agent workflows, and document-heavy pipelines without locking into a closed API. It pairs a 1-million-token context window with MIT-licensed weights you can self-host or call through more than 20 providers.
Core Features
- A Mixture-of-Experts transformer with 744 billion total parameters and 40 billion active per token, so only a fraction of the network runs on each step.
- A 1,048,576-token context window, up from 200K on GLM-5.1, built on a sparse-attention indexer that cuts per-token compute by roughly 2.9x at long range.
- Two reasoning levels, function calling, structured JSON output, and context caching that reuses prompt prefixes to lower repeat-work cost.
- Broad deployment: first-party API plus DeepInfra, Fireworks, Novita, SiliconFlow, and OpenRouter, all OpenAI-compatible.
Use Cases
- Long-running agentic coding and software-engineering tasks where it leads open models on PostTrainBench and hits 1595 on the Frontend Code Arena.
- RAG and document-heavy research that benefits from a 1M-token window holding entire repositories or case files in one pass.
- Teams wanting open-weight flexibility for self-hosting, provider failover, or cost tuning instead of a single endpoint.
Pricing
The first-party API runs about 1.40 dollars per million input tokens, 0.26 dollars cached input, and 4.40 dollars output, while third-party providers range from roughly 0.93 to 3.00 dollars input and 3.00 to 10.25 dollars output per million tokens. Weights are free under MIT for commercial and noncommercial use.
Our Take
Best open-weight option for agentic coding and long-context work at a sane cost, and the clear leader among open models on Artificial Analysis Intelligence Index at 51. The catch is verbosity and contested reproducibility. More at aifreetool.site/tool-category/ai-engine-model/.
