GLM-5.2: Open-Weight Frontier Coding at a Fraction of the Cost
Z.ai's GLM-5.2 ships open weights under MIT and reportedly leads the open-weight field on SWE-bench Pro while trading blows with Opus 4.8 on FrontierSWE and MCP-Atlas, at roughly one-sixth of GPT-5.5's blended cost. The catch is a hosted-API data-residency caveat.
