Laghu: a 152M chat model trained from scratch for $146. Everything is open.
• 3
AI research from first principles.
AI research from first principles. Small open models that run on your own device. udgamlabs.com · GitHub
| Model | What it does | Try it |
|---|---|---|
| Laghu | 152M chat model trained from scratch for $146; fully open, with checkpoints, code and evals | browser · Space |
| Shruti | Hindi, English and Hinglish speech to text in 29.5 MB; built on Cactus Whistle | browser · Space |
| Gargi | 384M chat model that thinks step by step | browser |
| Lilavati | 95M maths solver: writes a short Python program and runs it | browser |
| Chitra-Tiny | 212M text-to-image, 256×256 | browser |
| Udgam-384M-Chat | On-device tool calling: app actions, document to JSON, shell commands |
GGUF, MLX and datasets are in each model's collection below. Early experiments: udgam-english-125m · hinglish-commands-0.6b (a Qwen3-0.6B adapter).
27 small language models on one harness, pinned
152M chat model trained from scratch for $146
29.5 MB offline speech-to-text for Hindi, English, Hinglish
95M model that solves maths word problems by writing Python
384M model trained from scratch that thinks step by step