VibeThinker-3B: A 3B Dense Reasoning Model Built on Qwen2.5-Coder-3B With the Spectrum-to-Signal Post-Training Pipeline

Short summary

VibeThinker-3B is a 3-billion-parameter open-source reasoning model built on Qwen2.5-Coder-3B with a Spectrum-to-Signal post-training pipeline. It claims to match DeepSeek V3.2 and Kimi K2.5 performance on verifiable benchmarks. The MIT-licensed model is available for research and production deployment.

• •3B parameter reasoning model based on Qwen2.5-Coder-3B
• •Claims to match DeepSeek V3.2 and Kimi K2.5 benchmark performance
• •MIT-licensed open-source for research and commercial use

Generated with AI, which can make mistakes.