Before 2026, What will be true of OpenAI's Claimed IMO Gold Performance?
8
Ṁ256
Dec 31
90%
At least one former IMO medalist will review the model's answers and claim it did not actually achieve Gold
66%
The model that achieved it was trained with a new reinforcement learning algorithm
50%
The model that achieved it could earn at least bronze using no more than 100,000 reasoning tokens per question
41%
It was achieved with the same model OpenAI used to get second place in AtCoder
34%
The breakthrough is mostly the result of superior test-time scaling methods
34%
I will consider the techniques used to achieve it at least as big of a breakthrough as strawberry
24%
It was achieved by a model that does not use a standard transformer architecture

  • Update 2025-07-21 (PST) (AI summary of creator comment): A claim will only be considered valid if the person making it can prove to the creator that they are a former IMO medalist.

  • Update 2025-07-24 (PST) (AI summary of creator comment): The creator has specified what constitutes proof of being a former IMO medalist. Acceptable methods include:

    • Self-doxxing to the creator.

    • Posting a confirmation on a personal social media account.

#️ Technology
#AI
#OpenAI
#Technical AI Timelines
#IMO Grand Challenge
Get Ṁ1,000 play money
4 Comments
Sort by:
At least one former IMO medalist will review the model's answers and claim it did not actually achieve Gold
bought Ṁ50 At least one former ... YES

@JaundicedBaboon I claim the model did not actually achieve gold. Resolves yes.

@DottedCalculator Can you prove you're an IMO medalist?

@JaundicedBaboon How can I prove this?

@DottedCalculator Self-doxx or just write something on a personal social

Related questions