# WeiboAI/VibeThinker

Tiny Model, Big Logic: Diversity-Driven Optimization Elicits Large-Model Reasoning Ability in VibeThinker-1.5B

Repository: https://github.com/WeiboAI/VibeThinker
Canonical: https://ross.abutalabs.com/products/vibethinker
Language: Python
License: MIT
License Family: permissive
Topics: ai, huggingface, language-model, transformer, aime2025, livecodebench, reasoning-language-models, reasoning-models, sllm, llm
Last push: 2026-08-14T15:11:41+00:00

## Health v2 (maintenance only)
Score: 60/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 97, release rhythm 35, longevity 21
- inputs: {"age_days": 302, "days_push": 19, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1561, forks 116 (observed 2026-08-28T04:05:03.853596+00:00)

## What it is
VibeThinker is a family of small dense reasoning language models (1.5B and 3B parameters) from WeiboAI, trained with a Spectrum-to-Signal post-training pipeline to achieve frontier-level math and coding reasoning. The repository hosts documentation, papers, and links to model weights on Hugging Face and ModelScope.

## Use cases
- run a small reasoning model for math olympiad problems
- download a 1.5B LLM that rivals much larger reasoning models
- study how diversity-driven post-training elicits reasoning in tiny models
- benchmark small models on AIME and LiveCodeBench
- deploy a low-cost reasoning LLM locally
- research small-model reinforcement learning pipelines

## When to choose
- you need strong math/competitive-programming reasoning from a tiny model that fits on modest hardware
- you want to study or reproduce the Spectrum-to-Signal post-training approach
- you need open MIT-licensed reasoning model weights

## When to avoid
- you need general-purpose chat, long-context, or multimodal capabilities
- you require frontier-scale knowledge breadth rather than verifiable reasoning tasks
- you want a ready-made inference server or application rather than model weights

## Facets
- artifact type: learning-resource
- maturity: active
- function: machine-learning, llm-training, llm-inference
- domain: large-language-models, machine-learning, artificial-intelligence
- platform: python
- tags: reasoning-models, small-language-models, post-training, reinforcement-learning, math-reasoning, model-weights, huggingface, gpu

## Member repositories
- WeiboAI/VibeThinker (main) score 60

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:03.853596+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:29:43.697751+00:00, confidence not recorded.
  - readme: https://github.com/WeiboAI/VibeThinker (fetched 2026-08-28T04:05:03.853596+00:00, sha 56c96955c3e7)
- Data as of 2026-08-30T08:39:29.467469+00:00.