Did we ever learn what model won gold at the IMO from OpenAI? It was a year ago and it was called an unreleased internal general purpose model back then. Has GPT-5.5 Pro Extended caught up with whatever it was?
OpenAI's unreleased internal general-purpose model won gold at last year's International Mathematical Olympiad but remains publicly unidentified with no name or details released
Opacity also surrounds models used in Figure 01 demos.
Users highlight that current GPT-5 variants like 5.5 pro have solved IMO-level problems on MathArena, showing optimism about rapid progress in advanced math capabilities.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@emollick Was never named as far as I know. We also never learned the name of the OAI model that was in Figure 01.
Did we ever learn what model won gold at the IMO from OpenAI? It was a year ago and it was called an unreleased internal general purpose model back then. Has GPT-5.5 Pro Extended caught up with whatever it was?

@emollick "GPT-5.5 recently solved the last unsolved problem on MathArena Apex (the infamous IMO Problem 6), and Apex Shortlist is now also at 90%+ accuracy" https://matharena.ai/no_final_answer/

Yes 5.2 is glimpse and 5.4 got 5 of 6 Q right. Per http://matharena.ai Blog post and under apex Q12, also Q6 can be intermittently solved by GPT-5.5 xhigh (i am sure pro would solve it more often or always right). So we are better now on IMO and can see IMO 2026 goes. USAMO 2026 humans did really bad compared to GPT-5.5 98%

@emollick GPT 5.5 pro would score higher on IMO than that model. GPT 5.2-5.5 xhigh/pro are all capable of IMO gold

@emollick Would love to know

@emollick Wouldn’t it be one of the Pro models?

@emollick blackwell