VibeThinker-1.5B - 1.5 billion parameter large-scale language model open-sourced by Weibo AI

What is VibeThinker-1.5B?

VibeThinker-1.5B is a 1.5 billion parameter large-scale language model open-sourced by Weibo AI. Fine-tuned based on Alibaba's Qwen2.5-Math-1.5B, it is optimized for mathematical and coding tasks, with outstanding performance and industry-leading inference performance. The model adopts the Spectral-Signal Principle (SSP) training framework, which is divided into two phases: supervised fine-tuning and reinforcement learning, and amplifies signals by optimizing paths so that even small models can efficiently explore the inference space.

Features of VibeThinker-1.5B

• Strong math and coding skills: VibeThinker-1.5B excels at mathematical reasoning and code generation tasks, and after fine-tuning, its inference performance is industry-leading, outperforming even models with larger parametric counts.
• Efficient training framework: Adopting the Spectral-Signal Principle (SSP) training framework, we divide supervised fine-tuning and reinforcement learning into two phases, focusing on diversity first, and then optimizing the path through reinforcement learning to achieve signal amplification, so that even small models can efficiently explore the inference space.
• Low-cost training: The post-training cost of the model is only $7,800, which is much lower than the hundreds of thousands or even millions of dollars required for similar or larger-scale models, making it more cost-effective.
• Open Source and Free Access: VibeThinker-1.5B is now available for free on platforms such as Hugging Face, GitHub, and ModelScope, following the MIT license, and is available to researchers and developers for free, even for commercial purposes, lowering the development barrier.
• Multi-platform support: The model supports a variety of mainstream platforms, making it easy for users to choose the right environment for development and deployment according to their needs, with a high degree of flexibility and scalability.

Core Benefits of VibeThinker-1.5B

• High-performance reasoning capabilities: Demonstrate superior inference performance on mathematical and coding tasks, and can efficiently handle complex problems, outperforming models with partially larger number of parameters.
• Low-cost training and optimizationThe innovative "Spectrum-Signal Principle" training framework effectively reduces training costs, with post-training costs of only $7,800, significantly lower than the industry average.
• Open Source and Free Access: Following the MIT open source protocol, it is freely available to researchers and developers, supports commercial use, and greatly reduces the threshold and cost of use.
• Multi-platform support and ease of use: The model is live on multiple platforms such as Hugging Face, GitHub, and ModelScope, facilitating rapid deployment and use with good compatibility and ease of use.
• Efficient training framework: Optimizing model performance through a combination of supervised fine-tuning and reinforcement learning enables even small models to achieve efficient reasoning, improving the utility and flexibility of the model.

What is the official website for VibeThinker-1.5B?

• GitHub repository:: https://github.com/WeiboAI/VibeThinker
• Hugging Face Model Library:: https://huggingface.co/WeiboAI/VibeThinker-1.5B
• arXiv Technical Paper:: https://arxiv.org/pdf/2511.06221

People for VibeThinker-1.5B

• Artificial intelligence researchers: Its open source code and model architecture can be used for academic research to explore optimization and innovation in language models.
• Developers and engineers: For application development that requires efficient reasoning and code generation that can be quickly integrated into projects to improve development efficiency.
• data scientist: It can be used for data analysis and processing tasks to optimize data processing processes with its powerful mathematical reasoning capabilities.
• Educators and students: Provide practical tools for teaching machine learning and artificial intelligence to help students better understand and practice the techniques.
• business user: Suitable for enterprises that need low-cost, high-performance AI solutions that can be used for commercial application development to drive intelligent business transformation.