High-Quality Datasets
We generate reasoning datasets by querying frontier models with high reasoning effort settings, capturing their step-by-step thought processes.
Welcome
About us
TeichAI is focused on one mission: making frontier AI capabilities accessible through open-source model distillation.
We take the reasoning capabilities of proprietary models like GPT-5, Claude 4.5 Opus, Gemini 3 Pro Preview, and DeepSeek v3.2, and transfer that knowledge to smaller, open-source models that you can run locally on your own hardware.
What is Model Distillation?
Knowledge distillation is a technique where a smaller “student” model learns to replicate the behavior of a larger “teacher” model. In our case:
- Teacher Models: Frontier proprietary models (GPT-5, Claude 4.5, Gemini 3, etc.)
- Student Models: Open-source base models (Qwen3, Nemotron, etc.)
- Knowledge Transfer: We query the teacher models to generate high-quality reasoning traces, then fine-tune the student models on these outputs
┌─────────────────────────────────────────────────────────────┐│ DISTILLATION PIPELINE │├─────────────────────────────────────────────────────────────┤│ ││ ┌──────────────┐ ┌──────────────┐ ││ │ Teacher │ Query │ Dataset │ ││ │ (GPT-5, │ ──────► │ (Reasoning │ ││ │ Claude) │ │ Traces) │ ││ └──────────────┘ └──────┬───────┘ ││ │ ││ ▼ ││ ┌──────────────┐ ┌──────────────┐ ││ │ Student │ SFT │ Distilled │ ││ │ (Qwen3, │ ◄────── │ Model │ ││ │ Nemotron) │ │ │ ││ └──────────────┘ └──────────────┘ ││ │└─────────────────────────────────────────────────────────────┘Our Approach
Efficient Training
Using Unsloth and LoRA, we can fine-tune models efficiently on consumer GPUs (as little as 16GB VRAM for 4B models).
Multiple Formats
Every model is released in both HuggingFace Transformers format (16-bit) and GGUF quantizations for local deployment.
Full Transparency
All datasets, training scripts, and model weights are publicly available. You can replicate everything we do.