Is it worth paying $200 per month for coding help from Opus? Engineers are starting to wonder, prompted by the release of GLM-5.2.
China's Z.ai launched the MIT-licensed, 744-billion parameter Mixture-of-Experts text-only model, with a 1M-token context window, on June 16.
Two weeks later, cloud platform Databricks is seeing "astonishing" demand for GLM-5.2. That's according to a member of its technical staff, Yuchen Jin, who described it as "the open-source Claude moment" over the weekend.
In a rapid ascent, GLM now ranks seventh in OpenRouter's usage ranking, with the company declaring it is the "open model that makes Opus-style agentic coding portable".
A growing list of organisations have voiced that GLM5.2 could be, in effect, a cheaper version of Claude – and the similarities have raised suspicions that the Chinese model may be distilled from the frontier predecessor.
Opus, but cheap
Last week, application security platform Semgrep published the results of a scanning exercise that pitted the top open weight models against Claude. To its considerable surprise, GLM-5.2 beat Claude Code by a significant margin.
