Problem

When a user creates a language resource with a model that does not support the chat completions endpoint (e.g. gpt-5.4-pro), TQE analysis fails on every API request with HTTP 404: "This is not a chat model and thus not supported in the v1/chat/completions endpoint." The system does not recognize this as a permanent error, so it retries each segment up to the configured max retries. In the batch path, failed batches also fall back to sequential per-segment requests, all of which fail again. This results in a large number of pointless API calls and delays before the worker finally gives up. 

Solution

Detect and handle this kind of errors and prevent TQE for following segments.