[Suggestion / Feature] Allow custom model API keys in chat.z.ai + use the Coding Plan endpoint for prioritized model access
Observed context (from community discussion, not official docs):
- chat.z.ai today exposes Z.AI's own GLM models. Users frequently ask how to bring their own keys / use non-GLM models, but there is currently no in-UI option to add a custom endpoint or API key inside chat.z.ai.
- The Coding Plan API endpoint is real and already used by the community — users route it into external tools (Claude Code, Zed) to get prioritized/plan-rated access to models. That same prioritized path is not available from inside chat.z.ai itself.
Proposal:
- Custom model API support in chat.z.ai. Let users add their own API keys / base URLs (OpenAI-compatible, Anthropic, or any endpoint) so they can use non-GLM models alongside GLM in the same chat UI.
- Coding Plan endpoint inside chat.z.ai. Expose the Coding Plan API endpoint as a selectable backend in chat.z.ai, so users get the plan's prioritized access to all models without leaving the web UI or wiring up external tools.
Why this matters — especially for overload:
During GLM-5.2 saturation (the [1305] overload loop), GLM models are unavailable but a user's own API key (or the Coding Plan endpoint's prioritized path) may still work. Letting users fail over to:
- their own non-GLM key, or
- the Coding Plan endpoint's prioritized access
...turns a hard "model overloaded, reconnecting…" wall into a graceful continuation. It also reduces load on the public GLM pool by letting heavy users offload to their own endpoints.
Why this matters: Giving users a built-in offload path during peaks improves uptime perception and keeps work flowing, without Z.AI adding servers.
Regards,
Roman
[Suggestion / Feature] Allow custom model API keys in chat.z.ai + use the Coding Plan endpoint for prioritized model access
Observed context (from community discussion, not official docs):
Proposal:
Why this matters — especially for overload:
During GLM-5.2 saturation (the
[1305]overload loop), GLM models are unavailable but a user's own API key (or the Coding Plan endpoint's prioritized path) may still work. Letting users fail over to:...turns a hard "model overloaded, reconnecting…" wall into a graceful continuation. It also reduces load on the public GLM pool by letting heavy users offload to their own endpoints.
Why this matters: Giving users a built-in offload path during peaks improves uptime perception and keeps work flowing, without Z.AI adding servers.
Regards,
Roman