Zhipu's GLM 5.2 matches Opus 4.8 within one point and takes Anthropic's enterprise traffic at a fifth of the cost
Zhipu's GLM 5.2 — MIT-licensed, full open weights released June 16 — lands within one percentage point of Opus 4.8 on FrontierSWE and posts Intelligence Index 51 (the highest open-weights score Artificial Analysis has recorded), priced at $1.40 input / $4.40 output per million tokens against Opus 4.8's $5 / $25; OpenRouter traffic for the model is climbing faster than it did after DeepSeek's V4 launch in April. The US Commerce Department's June 12 order forced Anthropic to disable Fable 5 and Mythos 5 globally the day before GLM 5.2 reached subscribers — two of three US frontier labs now operate under government release gates while Zhipu, Moonshot, and Alibaba ship open weights enterprises can self-host and own outright, though Z.ai's cloud API remains subject to China's National Intelligence Law.
Source: cryptopolitan.com ↗
GLM 5.2 is the first open model that can compete with closed-source models.
Why this matters
- → Enterprise AI budgets shift to China as US export controls restrict Anthropic access
- → Open-weights model eliminates vendor lock-in; organizations retain capability offline
- → Performance parity at one-fifth cost accelerates adoption despite security concerns