GLM-5.2 is Z.ai's flagship open-weight model for long-horizon coding and agentic work. It is a text-only Mixture-of-Experts model with 753B total parameters and approximately 40B active parameters, released under the MIT License.

Use GLM-5.2 in LM Studio Bionic

GLM-5.2 is available as a Cloud model in LM Studio Bionic, with up to 256K tokens of context in the current LM Studio Cloud configuration.

Built for long-horizon engineering

Z.ai trained GLM-5.2 for sustained work across large codebases, complex debugging, research reproduction, and multi-step tool use. The model architecture supports up to 1M tokens of context and multiple reasoning effort levels, letting users trade latency and cost for more computation on difficult tasks.

GLM-5.2 introduces IndexShare, which shares one indexer across every four sparse-attention layers. Z.ai reports that this reduces per-token FLOPs by 2.9× at 1M context. Changes to its multi-token prediction layer also increased speculative-token acceptance length by up to 20% in Z.ai's ablation.

Benchmark results

Z.ai reports substantial gains over GLM-5.1 across coding and long-horizon evaluations:

Benchmark	GLM-5.2	GLM-5.1	Claude Opus 4.8
SWE-bench Pro	62.1	58.4	69.2
Terminal-Bench 2.1 (Terminus-2)	81.0	63.5	85.0
FrontierSWE	74.4	30.5	75.1
PostTrainBench	34.3	20.1	37.2
MCP-Atlas (Public Set)	76.8	71.8	77.8

Benchmark image and results are from Z.ai's GLM-5.2 launch post. FrontierSWE and PostTrainBench used 1M context, max reasoning effort, and up to 128K output tokens. Terminal-Bench 2.1 used the Terminus-2 harness with 256K context. See the launch post for the complete table and methodology.

Sources and license

GLM-5.2 is released under the MIT License. See Z.ai's launch post, model documentation, and official model card for more details.