VibeThinker-3B: Exploring the Frontier of Verifiable Reasoning in Small Language Models

Benchmark Model Rank Results
instruction-following-on-ifbenchVibeThinker-3B#13Accuracy: 74.5
code-generation-on-livecodebench-v6VibeThinker-3B#19Pass@1: 80.2
multiple-choice-question-answering-mcqa-on-gpqa-diamondVibeThinker-3B (CLR)#79Accuracy: 72.9