glm-5.1

GLM-5.1 was released and open sourced on April 8, 2026. It adopts MoE sparse architecture, with a total parameter of 744B, an activation parameter of about 40B, a context window of 200K, and a maximum output of 128K. Its landmark breakthrough lies in its long-range mission capability: the model can work independently and continuously for more than 8 hours in a single mission, during which it plans, executes, and evolves independently, and finally delivers complete engineering-level results. In the actual test, GLM-5.1 took 8 hours to execute more than 1,200 steps and built a fully functional Linux desktop system from scratch. On SWE-bench Pro, GLM-5.1 topped the world with a score of 58.4, surpassing GPT-5.4 and Claude Opus 4.6; Terminal-Bench 2.0 scored 63.5. The model is open sourced under the MIT license, launched on Huawei Cloud on the day of release, and can be directly called through tools such as Claude Code.

Context

1024000 tokens

Max output

131072 tokens

Input

llm

Output

llm