# NovaSky-AI/SkyThought

Sky-T1: Train your own O1 preview model within $450

Repository: https://github.com/NovaSky-AI/SkyThought
Canonical: https://ross.abutalabs.com/products/skythought
Homepage: https://novasky-ai.github.io/
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2025-07-12T03:05:27+00:00

## Health v2 (maintenance only)
Score: 25/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 31, release rhythm 8, longevity 42
- inputs: {"age_days": 601, "days_push": 417, "days_rel": 558, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3399, forks 344 (observed 2026-08-28T04:08:02.769324+00:00)

## What it is
SkyThought is the open-source repository behind Sky-T1, a family of reasoning language models trained for under $450, including training scripts, data curation, and evaluation tooling. It also hosts Skythought Evals, a Python package for evaluating and generating data for reasoning models, plus test-time scaling and RL training experiments.

## Use cases
- train your own o1-style reasoning model cheaply
- evaluate reasoning models on math and code benchmarks
- generate training data for reasoning model fine-tuning
- run test-time scaling for code generation
- fine-tune an llm with reinforcement learning
- reproduce the sky-t1 training pipeline

## When to choose
- you want to train or fine-tune a reasoning model on a budget
- you need reproducible evaluation harnesses for reasoning benchmarks
- you want to experiment with RL or distillation for LLM post-training

## When to avoid
- you just want to serve or chat with a model rather than train one
- you need a production inference framework
- you lack GPU resources for large-model training

## Facets
- artifact type: library
- maturity: active
- function: llm-training, machine-learning, benchmarking, data-generation, rag
- domain: large-language-models, deep-learning, machine-learning, artificial-intelligence
- platform: python, cloud
- tags: reasoning-models, reinforcement-learning, distillation, test-time-scaling, post-training, open-models, sky-t1, gpu, linux

## Member repositories
- NovaSky-AI/SkyThought (main) score 25

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:02.769324+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 2026-08-29T18:38:46.389799+00:00, confidence not recorded.
  - readme: https://github.com/NovaSky-AI/SkyThought (fetched 2026-08-28T04:08:02.769324+00:00, sha f3cb3456a9fb)
  - homepage: https://novasky-ai.github.io/ (fetched 2026-08-29T09:32:56.613031+00:00, sha 48efb8f1d377)
  - registry_pypi: https://pypi.org/pypi/skythought/json (fetched 2026-08-29T09:32:56.622146+00:00, sha 7842409aa5d7)
- Data as of 2026-08-30T08:39:29.467469+00:00.