Zhipu Releases GLM-5.2: 1M Token Lossless Context, Open-Source SOTA in Long-Horizon Coding
On June 13, Zhipu AI officially released GLM-5.2, its new flagship open-source large language model. This marks Zhipu's third major iteration in just four months, with core breakthroughs focused on ultra-long context handling and long-horizon coding agent capabilities, achieving open-source SOTA on multiple benchmarks.
Core Upgrade: 1M Token Lossless Context
The headline feature of GLM-5.2 is support for a 1 million token lossless context window. In long-horizon tasks, the model significantly reduces "context drift" and "goal forgetting" — the two most common failure modes for LLMs handling complex multi-step tasks. For scenarios requiring sustained tracking of large codebases, long-document analysis, or multi-turn complex reasoning, this upgrade delivers a qualitative leap in stability.
Long-Horizon Coding Agent Specialization
Zhipu revealed that the team conducted months of specialized training for long-horizon Coding Agent scenarios, covering large-scale code implementation, automated research, performance optimization, and complex debugging. On authoritative Coding and long-horizon task benchmarks, GLM-5.2 achieves open-source SOTA, with particularly stable performance in complex systems engineering and deep debugging scenarios.
Architecture & Parameters
GLM-5.2 continues the MoE (sparse mixture-of-experts) architecture with Dynamic Sparse Attention (DSA), totaling 744B parameters with ~40B activated. Training data extends to November 2025. The current release focuses on text and code.
Flexible Reasoning Effort Control
The newly introduced multi-tier reasoning effort control is a standout feature, allowing users to freely balance model capability against execution speed and compute cost. At equivalent token consumption, GLM-5.2's agent programming performance significantly outperforms previous GLM generations.
Open Source & Availability
GLM-5.2 is open-sourced under the MIT license, available to all GLM Coding Plan users (Lite/Pro/Max/Team), with API access launching the week after release. This means developers and enterprises can integrate this top-tier model into their products at zero cost.
Industry Impact
GLM-5.2's release intensifies competition in the open-source LLM space. With 1M token context and open-source SOTA coding capabilities, it positions itself as a strong competitor to underlying models in tools like Claude Code and Cursor. For developers focused on AI programming efficiency, this is a major update worth watching closely.
Want to see how GLM-5.2 compares against other leading coding models? Visit our AI Tool Comparison Center.