glm-5.2

GLM-5.2 was officially released and open sourced on June 17, 2026. It is a new generation flagship model of the GLM-5 series and is designed for long-cycle software engineering tasks. The model continues the MoE route, with a total parameter of 744B, activation parameters of about 40B, the context window has jumped from 200K in 5.1 to 1 million Tokens, and the maximum output is 131K. Key technical improvements include the IndexShare mechanism, which reduces the amount of calculation per Token by 2.9x under millions of contexts. On the FrontierSWE benchmark, GLM-5.2 scored 74.4, just about 1 point shy of Claude Opus 4.8; Terminal-Bench 2.1 jumped from 63.5 to 81.0. The model is open sourced under the MIT protocol and has been adapted to eight major domestic computing platforms including Huawei Ascend and Cambrian on the first day of launch.

Context

1024000 tokens

Max output

131072 tokens

Input

llm

Output

llm