GLM 5.2

GLM 5.2, released by Tsinghua-affiliated lab Z.ai in June 2026, is an advanced open-weight model positioned as a long-horizon agentic powerhouse. It was released 11 weeks following its predecessor, GLM 5.1, and includes a 5x context expansion alongside a new architecture.

Architecture and Capabilities

GLM 5.2 is specifically designed to handle long-context, long-session agentic work stably and efficiently. The model features an IndexShare architecture that reuses one lightweight indexer across every four sparse-attention layers. This cuts per-token FLOPs by roughly 2.9x at a 1M context. It also includes an improved MTP layer for speculative decoding extending the accepted token length by up to 20%.

Benchmark Performance

Z.ai explicitly targeted enterprise developers running long, complex coding sessions on large codebases. On SWE-bench Pro, GLM 5.2 scores 62.1%, outperforming GPT-5.5's 58.6%. On FrontierSWE—a benchmark testing sustained multi-step reasoning across full agentic sessions—it achieved 74.4%. It scored 51 points on the Artificial Analysis Intelligence Index for open-weight models, placing it at the top globally at the time of release.

Pricing and Tradeoffs

While GLM 5.2 leads the open-weight pack in production coding benchmarks, its output pricing is slightly higher than open competitors at USD 4.40 per million output tokens—although this is still 5.7 times cheaper than models like Claude Opus 4.8. For developers who need focused, robust performance over long agentic software refactors instead of generalized multimodality, GLM 5.2 offers a powerful choice.