Skip to main content

​
Chat & model troubleshooting

Use this guide when a conversation will not send, a model is missing, streaming stops, or a capability behaves differently between models.

​
Model missing

Check:
  1. platform model state
  2. plan entitlement
  3. active organization policy
  4. user/resource override
  5. staged rollout
  6. provider health
A different model list between two users is not automatically a bug.

​
Message will not send

Confirm:
  • session is still valid
  • selected model remains available
  • required attachment finished uploading
  • model supports attached input type
  • tool/workload is allowed by policy
  • credit/hard limit is not exhausted
Try a new conversation and another eligible model to narrow scope.

​
Streaming stops

A dropped client connection does not always mean the provider/run failed. Check whether the response completed server-side when the product exposes that state. If failures repeat across unrelated chats, inspect provider/system health.

​
Tool call fails

Separate:
  • model cannot use tools
  • tool not enabled for this context
  • tool permission denied
  • tool provider/runtime failed
  • tool timed out
Do not switch to a model without tool support and assume the request is equivalent.

​
Context too large

Reduce unrelated history/attachments, start a new conversation, or move durable information into Project/Knowledge retrieval.

​
Model behavior changed

Check whether the selected model/route changed, model version/config changed, reasoning mode changed, or routing/fallback selected a different eligible route.

​
Broad failure

If many users and models fail simultaneously, use System Health and Incidents rather than applying per-user overrides.

Model capabilities

Compare the capabilities required by different chat workloads.