Model
gpt-5-thinking-mini
A small model variant in OpenAI's GPT-5 series with chain-of-thought reasoning capability, which tends to refuse on uncertain questions in order to reduce the error rate.
Timeline (last 90 days)
Oct 18
On the SimpleQA benchmark, gpt-5-thinking-mini declines to answer 52% of questions, with an error rate of 26%.
Unverified50%
Oct 18
o4-mini and gpt-5-thinking-mini have almost the same accuracy on SimpleQA, but the latter makes roughly three times fewer errors than the former.
Unverified50%