AIMarkTechPostabout 3 hours ago
VibeThinker-3B matches giant models on reasoning tasks
6 min read
Researchers from Sina Weibo Inc released VibeThinker-3B, a 3-billion-parameter dense reasoning model built on Qwen2.5-Coder-3B. Post-trained using the Spectrum-to-Signal pipeline, it scores 94.3 on AIME26, matching DeepSeek V3.2 (671B) and Kimi K2.5 (1T). The open-source MIT-licensed model runs on a single GPU.
Level
Hype check
Tap to vote and see what everyone thinks.
