WeiboAI

weights

open(MIT)

data

described(the card describes SFT, RL and self-distillation stages but releases no training data)

license

MIT(the model repository)+Qwen-Research-License(inherited from the Qwen2.5-Coder-3B base, non-commercial)

base

open_weights(Qwen2.5)

The weights ship under MIT, but VibeThinker is fine-tuned from Qwen2.5-Coder-3B, whose Qwen Research License grants rights for non-commercial purposes only and asks commercial users to request a license. That restriction travels with the weights to the fine-tune, so commercial use is not available. The card describes the training stages but releases no training data.

• https://huggingface.co/api/models/WeiboAI/VibeThinker-3B recorded 2026-10-08

  Hub metadata carries the tags "base_model:finetune:Qwen/Qwen2.5-Coder-3B" and "license:mit", reports the model as not gated, and declares no dataset on the card.

• https://huggingface.co/Qwen/Qwen2.5-Coder-3B/raw/main/LICENSE recorded 2026-10-08

  The base model's license is the "Qwen RESEARCH LICENSE AGREEMENT", which grants rights "FOR NON-COMMERCIAL PURPOSES ONLY" and says "If you are commercially using the Materials, you shall request a license from us."

• https://huggingface.co/WeiboAI/VibeThinker-3B/raw/main/README.md recorded 2026-10-08

  Model card: "The model repository is licensed under the MIT License." Its training section describes "Curriculum-based two-stage SFT" and "Multi-domain Reasoning RL" without linking any training data.

Adoption is measured as Hugging Face downloads on VibeThinker-3B, the product's one repository.

• https://huggingface.co/api/models/WeiboAI/VibeThinker-3B recorded 2026-10-07

  4,573 downloads in the trailing 30 days for WeiboAI/VibeThinker-3B

• https://huggingface.co/WeiboAI/VibeThinker-3B recorded 2026-08-13

  active HF repo; 822 likes; early traction as a small reasoning specialist

VibeThinker-3B reaches frontier-range results on narrow, verifiable math and code reasoning despite having only 3B parameters, according to its model card. It is explicitly not built for general or agentic use, so it sits below general-purpose frontier models.

• https://huggingface.co/WeiboAI/VibeThinker-3B recorded 2026-09-29

  Model page embeds eval results for MathArena/aime_2026 and Idavidrein/gpqa (diamond) with value 70.2; the page states "VibeThinker-3B reaches 76.4 on IMO-AnswerBench".

• https://huggingface.co/WeiboAI/VibeThinker-3B/raw/main/README.md recorded 2026-09-29

  Model card: "VibeThinker-3B reaches 76.4 on IMO-AnswerBench, a highly challenging benchmark with 400 IMO-level problems, with only 3B parameters"; recent LeetCode contests "corresponding to a" 96.1% acceptance rate; warns it "was not trained on tool-calling or agent-based programming data".