OpenBMB
Tiny, capable models for text, vision, audio and omni — small enough to live on your own hardware.
01What they offer
The MiniCPM family proves you don’t need a giant to get real work done. Each model is tuned to punch far above its parameter count and runs happily on a laptop — or a phone. OpenBMB provide free hosted API access for the jam, and every model also runs locally via llama.cpp or transformers. Pick the modality you need and go.
02Models & products
The kit, sized small
Full-duplex omni model — voice, vision and language in, speech out. Real-time capable.
03Build this with us
If you want to build…
Image understanding, OCR or documents
MiniCPM-V 4.6Vision + OCR
A video-understanding demo
MiniCPM-V 4.6 / 4.5Multimodal video
An omni or voice + vision assistant
MiniCPM-o 4.5Speech in, speech out
A lightweight, local-first text app
MiniCPM5-1B1B, on-device
Text reasoning / problem-solving
MiniCPM4.1-8BBest reasoning
TTS or a creative voice experience
VoxCPM2Voice generation
04Getting started
STARTER SPACE
MiniCPM-V-4.6 Demo
Fork this Gradio Server Space to start from a working MiniCPM-V-4.6 app.
05Eligible prize
OPENBMB · SPONSOR PRIZE
Best MiniCPM Build
To qualify ·Build with MiniCPM models. The pool is split $5k per track (1st $2,500 · 2nd $1,500 · 3rd $1,000).
$10,000
All prizes