Open Source AI Research

Distilled models and curated datasets for the community

We fine-tune open-source models on high-quality reasoning datasets from frontier models like Claude, GPT, and Gemini. All models are released in GGUF format for local deployment.

74.8K

Downloads (last 30 days)

62

Models Released

19

Datasets Published

What We Do

Model Distillation

We train open-source base models (Qwen3, GPT-OSS) on reasoning traces from frontier models. This transfers capabilities while keeping models small and efficient.

Dataset Curation

We create high-quality reasoning datasets by querying models like Claude Opus, Gemini Pro, and GPT-5 with diverse prompts covering coding, math, and science.

GGUF Quantization

All models are released in GGUF format with multiple quantization levels (Q4, Q8, etc.) for use with llama.cpp on consumer hardware.

Support Our Work

We're two college students funding this research out of pocket. Dataset generation costs add up quickly - our Claude Opus dataset alone cost $52+ to create. If you find our models useful, consider supporting us.

Donate via PayPal