GLM-5.2

GLM-5.2 is Zhipu AI's latest open-weight large language model, a 744-billion-parameter system tuned for coding, multi-step reasoning, and long-horizon agentic tasks. Released June 13-14, 2026 under an MIT license, it ships with a 1M-token context window and is positioned to compete directly with closed frontier models from OpenAI, Anthropic, and Google — except you can download the weights and run them yourself.

The headline isn't just the benchmark numbers (Zhipu claims results competitive with, and on some coding and agentic evaluations ahead of, top proprietary models). It's the licensing. MIT is about as permissive as open weights get: no usage caps, no commercial restrictions, no revocable API terms. For teams that want a frontier-class coding model they fully control — fine-tuning, on-prem deployment, no data leaving their infrastructure — that combination is rare. Most "open" models at this tier carry restrictive community licenses; GLM-5.2 doesn't.

It's built for the agentic era specifically: long context plus tool-use orientation means it's aimed at coding agents that plan, call tools, and run for many turns rather than one-shot completions. You can try it free through the Z.ai chat interface, get 20M free tokens on API signup, or self-host the weights. The catch is the same one every massive open model carries — running 744B parameters locally is a serious hardware commitment, so for most users the hosted API or chat UI is the realistic path.