GLM-5.2 Released: 1M Context and Enhanced Coding Performance

Zhipu AI has launched its new flagship model, GLM-5.2. This update is especially optimized for performing complex, long-horizon tasks by providing a stable 1M token context window.

Key features include:

• 1M Context Support: Enables stable task performance while maintaining quality even in large-scale coding agent scenarios.
• IndexShare Architecture: By reusing the same indexer across every 4 sparse attention layers, it achieves a 2.9x reduction in FLOPs per token in a 1M context environment.
• Flexible Effort Level Control: Users can select from various compute levels to balance performance and latency.
• MIT License: Adopts a Pure Open policy, allowing anyone to use it without regional restrictions.

In terms of performance, GLM-5.2 recorded the best open-source model performance on major coding benchmarks such as FrontierSWE, PostTrainBench, and SWE-Marathon, demonstrating performance on par with or exceeding closed-source models like Claude Opus 4.8 and GPT-5.5.