OpenAI announced on Saturday that it had won the prestigious gold medal, and now, Google is claiming the same.
For their part, they used an «advanced version» of the DeepThink reasoning model, not a specially tuned model, that included parallel thinking — which can «explore and combine multiple possible solutions before giving a final answer.»
They also fed the model a host of “high quality” solutions to previous problems, and general advice as to how it should approach IMO problems, which could have given it a significant boost.
This model solved five out of six problems in the exam, achieving a score of 35 out of 42. This is precisely the same score that OpenAI got, and is just barely above the threshold for the gold medal.
Google says the model should be available to testers “soon” and then roll out to Ultra subscribers on Gemini, without giving any certain time frame.
Read more on the contest itself in Sunday’s OpenAI story, and read Google’s announcement here and their X.com thread.