The model is designed for code processing and agent-based tasks.

The official description and benchmarks have not yet been published. In the amateur AICodeKing chart, GLM-5 took third place in non-agent-based tasks and first place in agent-based tasks, beating Claude Opus 4.6.

GLM-5 was available for testing on OpenRouter for several days under the anonymous name Pony Alpha and received positive reviews. Now you can try it for free!