OpenAI wins Math Olympiad gold, a major milestone for AI development

The gold medal in the Math Olympiad was achieved using no tools, no internet and with a pure reasoning model.
The International Mathematical Olympiad gold medal has been long sought by the AI industry, as it is one of the most difficult cognitive competitions out there. (Picture: OpenAI)
They got the accomplishment with a «general-purpose reinforcement learning» model, and not a specialized single-purpose math tool — using pure, natural language generative AI.

OpenAI says it was achieved with «our latest experimental reasoning LLM:»

According to Alexander Wei at OpenAI, the model followed the same rules as the 670 humans lining up for the competition, which means it solved 5 of 6 exceedingly difficult problems within the allotted time of two sessions of 4.5 hours.

No tools, no internet
The exam allows no tools, no internet access, and only about 10%, or 67, of the human participants sitting for the exam managed to get a gold medal, writes Engadget.

The OpenAI model got 35 out of 42 points available, just barely above the barrier for winning the gold, according to the contest’s official web page.

Too soon for ChatGPT 5
ChatGPT 5 is coming «soon,» the Olympiad tweets say, but don’t expect this kind of performance from it. They won’t be releasing this model until «many months» from now, which is too soon for GPT 5, CEO Sam Altman confirms in his tweet on the issue:

«This is an experimental model that incorporates new research techniques we will use in future models,» he says.

Read more: Alexander Wei’s post onXx, some background on Engadget.