Yes, the original announcement for o3 and o4-mini: https://openai.com/index/introducing-o3-and-o4-mini/ https://openai.com/index/introducing-o3-and-o4-mini/ o

Yes, the original announcement for o3 and o4-mini: https://openai.com/index/introducing-o3-and-o4-mini/ https://openai.com/index/introducing-o3-and-o4-mini/ o3 scored 91.6 on AIME 2024. 83.3 on GPQA o4-mini scored 93.4, 81.4 GPQA Then, the new announcement https://help.openai.com/en/articles/6825453-chatgpt-release-notes#h_6c98aab643 https://help.openai.com/en/articles/6825453-chatgpt-release-... o3 scored 90 on AIME 2024, 81 GPQA o4-mini wasn't measured --- Codeforces is the same, but they have a footnote that they're using a different dataset due to saturation, but still have no grounding model to compare with