Is Z.ai Down?

If Prismix shows Z.ai as "Operational" but you're having issues:

• GLM-4 web search plugin returning Chinese results only — GLM-4's built-in web search plugin currently retrieves results primarily from Chinese-language search indexes. English-language web search is not supported in the web-search plugin mode. For multilingual retrieval, implement your own search tool using a global search API (e.g., Brave Search, Serper) and pass results to GLM-4 via the function calling interface.
• API key from open.bigmodel.cn not working with z.ai web interface — Z.ai (the consumer chat at z.ai) and the BigModel developer API at open.bigmodel.cn are separate products with separate account systems. An API key generated at open.bigmodel.cn is for programmatic API access only and cannot be used to sign into z.ai's chat interface, which requires a standalone Zhipu account login.
• Visual model failing on images over 4 MB — GLM-4V and other visual variants of the GLM model family enforce a 4 MB per-image limit. Uploading larger images results in a payload too large error. Resize or compress images before sending them in API requests, targeting under 4 MB while preserving sufficient resolution for the analysis task.
• Streaming response not completing (connection drops at token limit) — GLM-4 streaming responses can be terminated abruptly when the generated output reaches the model's max token limit mid-stream. The stream ends without a proper stop token being sent. Increase the max_tokens parameter and set a longer read timeout in your HTTP client to ensure the full response is received.
• Free trial quota exhausted silently (check quota on dashboard) — Zhipu AI's free trial tokens are consumed without visible warnings in the API response. Once the trial quota is exhausted, API calls return a quota exceeded error. Monitor your remaining token quota on the BigModel dashboard at open.bigmodel.cn under account settings and add tokens or upgrade your plan before quota runs out.
• Model version not matching documentation (aliased to latest automatically) — model slugs like glm-4 are aliased to the latest stable release and can change behavior after model updates. If a new release introduces breaking changes in response format or capability, pin to a specific versioned slug (e.g., glm-4-0520) to lock your application to a known-good model version.