GLM-5.2 matches Opus quality, cuts costs

The results showcase the closing gap between open-weights and proprietary models, highlighting GLM-5.2 as a viable daily driver for high-volume developer workflows.

• –GLM-5.2 generated twice the number of tokens but remained faster and three times cheaper than Claude Opus, suggesting high inference efficiency.
• –The tests were based on real-world coding tasks, specifically building landing pages, where output quality was found to be functionally equivalent.
• –Releasing the prompt and cost evaluation suite allows the developer community to independently audit the performance claims.
• –Z.ai's MIT-licensed model represents a growing threat to closed-source API monetization, particularly for coding agent tasks.