Repeatedly, a request for GPT 6 Pro leads to ‘Thinking failed’ errors. In fact, this happens on more than 50% of my queries, and leads to hours of wasted effort.
This is extremely frustrating.
Has been happening for months.
This has been happening for me recently as well. Almost every prompt I send fails.
My last day to use up my 200 pro 6 messages, and everything is failing. Money back please
you can’t use 200 messages if they keep rate limiting you for even the ones that fail LOL
what a great way to save on compute! @OpenAI_Support !
There is now an official status incident that overlaps part of this thread, which gives us a useful timing boundary.
OpenAI Status recorded “Elevated Error Rates on GPT-6 Astra Pro” on September 24:
That overlaps the evening reports here around 19:12 and 19:18, so those failures should be treated as incident-confounded rather than clean evidence of some separate account-specific cause.
But it does not explain the whole report, because the original post says the “Thinking failed” problem has been happening for months and on more than half of queries.
So I would split the evidence into three windows:
The most useful next evidence is not another large batch of attempts. It is one naturally occurring post-resolution failure with:
If failures continue after the official recovery window, that separates the chronic issue from the known Sep-24 outage.
And I would keep “Thinking failed”, rate-limit accounting, and model-routing/substitution as separate questions unless native evidence actually ties them together.
