#40 · Primary category： Foundation Models
VibeThinker
Tiny Model, Big Logic: Diversity-Driven Optimization Elicits Large-Model Reasoning Ability in VibeThinker-1.5B
Project last updated：08/14/26
GitHub Stars
1.6K
Forks
117
Contributors
4
License
MIT
Why we included this project
VibeThinker is a good fit if you need serious math and coding reasoning from a model small enough to run on a modest GPU budget. It ships two dense models, 1.5B and 3B parameters, built on Qwen2.5 and post-trained to hold their own on verifiable reasoning tasks like competitive programming and math competitions. The technical reports claim the 1.5B model matches or beats far larger open-source systems on benchmarks such as AIME and HMMT, while the 3B release adds a broader training pipeline that includes test-time scaling. Weights are fully open with an MIT license, and the repo includes evaluation scripts plus the sampling settings the authors used, so you can reproduce the reported numbers. Just keep the scope in mind: these are specialized for answer-verifiable problems, not general open-domain knowledge, so treat them as a focused reasoning tool rather than a general assistant.
Articles for this project
No articles for this project yet.
To suggest a topic or contribute an article, contact us.
Related projects in this category
transformers
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.
CLIP
CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image
MiniCPM-V
A Pocket-Sized MLLM for Ultra-Efficient Image and Video Understanding on Your Phone
generative-models
Generative Models by Stability AI
unilm
Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities