Snowflake CEO Sridhar Ramaswamy ran 103 coding tasks and found that Chinese AI model GLM-5.2, made by Zhipu, nearly matches Anthropic's Claude Opus 4.7 in accuracy at a fraction of the cost. He shared the results on June 24, 2026. In the test, each task was run three times: GLM-5.2 solved 66% of tasks and Opus 4.7 solved 67%. On the first attempt, Opus got 53.7% right while GLM got 47.6% right. The cost difference is significant: GLM-5.2 charges $1.40 per 1M input tokens and $4.40 per 1M output tokens, compared to Opus 4.7's $5.00 and $25.00, respectively. However, GLM-5.2 used 860 million total tokens versus Opus 4.7's 439 million. Ramaswamy praised GLM for reliably checking code across both databases but flagged weaknesses: the model sometimes gave up on tasks too early and made too many tool calls. On one task, GLM made 411 tool calls over 24 minutes and still failed, while Opus solved it with 49 calls in 9 minutes.