Environment
- Surface: Web / iOS / Android / macOS / Windows/ Linux
- Browser and app versions
- Affected feature and model, if relevant: All allowed in my plan
Bug
subject: High/Extended reasoning in ChatGPT Chat is being routed to GPT-5.5-mini
Hello OpenAI Support,
I am reporting a persistent issue affecting regular ChatGPT Chat on my Plus account.
Since approximately September 4, 2026, manually selected Medium and High reasoning have frequently failed in regular ChatGPT Chat. Complex requests that previously triggered sustained reasoning, web research, source comparison, and detailed synthesis now often receive a shallow response almost immediately.
This issue affects regular ChatGPT Chat specifically. ChatGPT Work and Codex continue to work normally for me.
I have ruled out the most obvious explanations:
-
This is not a one-off response-quality issue; it has persisted for several days.
-
Retrying with “Try again” does not resolve it for me.
-
I have left the account unused for more than a day, but the problem remains.
-
I have not received a notice of a usage limit, model restriction, suspicious activity, or policy-related restriction.
-
The issue occurs even when I explicitly select High/Extended reasoning rather than using automatic routing.
I captured a response stream from an affected request in regular ChatGPT Chat on desktop web.
The client-side metadata at the beginning of the request states that I selected:
model_slug: gpt-5-6-thinking
thinking_effort: extended
However, the final server telemetry metadata for that same request states:
plan_type: plus
product_experience: chat
requested_model_experience: thinking
model_slug: gpt-5-5-mini
did_auto_switch_to_reasoning: false
is_autoswitcher_enabled: false
tool_invoked: true
turn_use_case: search
turn_mode: default
fast_convo: true
This appears to be a mismatch between the model and reasoning effort selected in the user interface and the model that actually served the request.
The request did invoke web search and generated several search queries, but it was treated as a fast Chat/Search turn and began producing visible output in about 3.2 seconds. This matches the actual symptom: despite selecting High/Extended reasoning, ChatGPT returns a short and low-effort response instead of performing sustained reasoning and comprehensive research.
Expected behavior:
-
Manually selecting High/Extended reasoning should use the selected reasoning model and allocate increased reasoning effort.
-
A complex research request should not be silently served by a mini model without clear user notification.
-
If a usage-limit fallback is involved, ChatGPT should show the relevant limit or reset information.
Could you please investigate whether my regular Chat experience is affected by an incorrect fallback, routing mismatch, entitlement-state issue, or a Chat-only backend regression?
I can provide screenshots of the selected reasoning level, the affected responses, timestamps, conversation links, and a sanitized HAR file if your technical team needs them.
For privacy and security, I have intentionally removed all authentication tokens, cookies, session/resume tokens, conversation IDs, request IDs, user text, and account identifiers from the metadata excerpt above.
Thank you