Z.ai’s GLM-5.2 launched in mid June with a 1M-token context window and two reasoning effort levels, according to DataCamp and the Ollama README. Tech Stackups ran a head-to-head test building a 3D platformer in raw WebGL and reports that Claude Opus completed the task in 33m 30s while GLM-5.2 took 1h 10m 40s, and Tech Stackups lists billed cost at $5.39 for GLM-5.2 versus ~$21.92 for Opus. Tech Stackups also reports Opus produced more output tokens and shipped a cleaner, faster result, while GLM-5.2 delivered comparable capability at lower cost and with open weights, per Tech Stackups and Ollama. Editorial analysis: For practitioners, the run illustrates a common tradeoff in agentic coding workflows between latency/cleanliness and cost/open-weight availability.
