I Ran the 3B Model That Beat Gemini 3 Pro at Olympiad Math — It Shouldn't Work

A 3-billion-parameter model from Weibo scored 94.3 on the AIME 2026 math Olympiad, outperforming Google's Gemini 3 Pro which scored 91.7. The model is MIT-licensed and its success challenges assumptions about model size and performance.

A 3-billion-parameter model just scored 94.3 on AIME 2026. Gemini 3 Pro scored 91.7. The 3B model is from Weibo, it is MIT-licensed, and… Continue reading on Towards AI »