Distill Frontier Models
We take the reasoning capabilities of models like GPT-5, Claude 4.5 Opus, Gemini 3 Pro, and DeepSeek v3.2 and transfer them to smaller, open-source models.
TeichAI Documentation
Bringing frontier AI capabilities to open-source models through knowledge distillation
What We Do
Open-Source Everything
All our models, datasets, and training scripts are freely available on Hugging Face and GitHub.
Run Locally
Our distilled models are designed to run on consumer hardware. Available in GGUF format for llama.cpp, Ollama, and LM Studio.
Transparent Process
We document every step of our distillation process so you can replicate our work or create your own distilled models.
Quick Links
📒 Notebooks
Ready-to-run Colab notebooks for training your own distilled models. View Notebooks →
🤗 Hugging Face
Browse our complete collection of models and datasets. Visit HF →
⚠️ SFT Limitations
Understanding the limitations of SFT-only distillation. Learn More →
🔧 Training Guide
Step-by-step guide to training parameters and configuration. Read Guide →