Weibo's VibeThinker-3B: A Game Changer in AI?

The Rise of VibeThinker-3B

A recent technical report from Weibo's research team has ignited debates in the AI community. Their model, VibeThinker-3B, has achieved remarkable scores on various prestigious math and coding benchmarks, raising questions about the validity of current AI scaling laws.

With a score of 94.3 on the AIME 2026, VibeThinker-3B is now competing with models that have hundreds of billions of parameters. This has led to skepticism among experts, who are questioning whether these benchmarks are still relevant or if they can be manipulated.

• Key achievements of VibeThinker-3B include:
• 91.4 on AIME 2025
• 80.2 Pass@1 on LiveCodeBench v6
• 93.4 on IFEval

The implications of these results could redefine the AI landscape, challenging the notion that larger models are inherently better. As the debate continues, the future of AI development hangs in the balance.