What's New in GLM 5.2. Run It on Featherless.

Z.ai's GLM-5.2, released on June 16, 2026, and hosted by Featherless from Day Zero, is a cutting-edge open-weight model excelling in coding benchmarks such as FrontierSWE, PostTrainBench, and SWE-Marathon, ranking alongside the likes of Claude Opus 4.8 and GPT-5.5. With a Mixture-of-Experts design and approximately 753 billion parameters, GLM-5.2 features a revised long-context architecture that enhances performance without increasing scale, focusing on coding agents. It introduces IndexShare and an improved multi-token-prediction layer to optimize long, complex sessions, offering a balance between reasoning depth and computational latency. In benchmarks, it performs competitively, trailing Claude Opus 4.8 slightly on some and surpassing GPT-5.5 on others, establishing itself as a leading model for open-weight coding tasks. Available through an OpenAI-compatible API with no regional limits or usage caps under an MIT license, GLM-5.2 can be integrated into commercial products and fine-tuned without restrictions. Featherless' hosting eliminates the need for GPU provisioning or inference-server setup, providing ease of access for developers.

No tracked trend matches for this post yet.