Home / Technology / GLM-5.2: Massive Open LLM Dominates Coding Tasks
GLM-5.2: Massive Open LLM Dominates Coding Tasks
17 Jun
Summary
- New GLM-5.2 LLM boasts 753 billion parameters and 1M token context.
- Released under MIT license, enabling free customization and local deployment.
- Outperforms leading proprietary models in crucial long-horizon coding benchmarks.
Chinese AI startup Z.ai has launched GLM-5.2, a 753-billion parameter open-weights large language model. This model is engineered for long-horizon autonomous coding and engineering tasks, featuring a 1-million-token context window. The core weights are released under an unrestricted MIT open-source license, allowing enterprises to freely download, customize, and potentially run the model locally.
GLM-5.2 introduces architectural optimizations like 'IndexShare' to reduce computational costs. It demonstrates state-of-the-art performance on various benchmarks, notably outperforming proprietary models such as OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8 in agentic tool use and long-horizon software engineering tasks.
The model is available through Z.ai's GLM Coding Plan and API, with competitive pricing significantly undercutting Western rivals. The open-weights approach and MIT license offer enterprises a path to deploy advanced AI locally, bypassing potential regulatory uncertainties and vendor lock-in associated with proprietary solutions.
