J
Julien
Hi there,
Thanks to the new “Auto” setting that automatically chooses the best language model. That's a so great feature!
However:
  1. After a request finishes, there’s no indication of which model was actually used...
  2. I can’t configure a block-list / allow-list to exclude certain models (for cost, latency, or compliance reasons).
  3. As a result, it’s hard to debug quality differences, reproduce results, or forecast cost, even if the answers are usually really good to me.
This leads to several impacts:
• It's totally impossible to trace surprising answers back to a specific model.
• It would be harder to maintain consistency in production workflows.
• It could lead to potential budget/quotas overruns if a "high-cost" model was silently chosen.
• It has a significant compliance risk: some projects must eventually avoid certain models whose...
That said, here are some suggestions:
  1. Surface the chosen model
• Show it in the UI (badge under the answer).
• Eventually return it in the API response header / JSON (
"model": "gpt-4o"
) for API users (if this is not the case, i can't check that)
  1. Propose a configurable allow / block list
• Let users exclude models globally in settings or per-request via a parameter.
• Alternatively, let us set a preferred order.
  1. Usage analytics
• Add a dashboard card summarizing how often each underlying model was selected by Auto mode over time.
This could clearly leads to some benefits:
• Transparent debugging and reproducibility.
• Controlled quotas and compliance.
• Greater trust in Auto mode, leading to wider adoption in professional workflows.