Model · OpenBMB · Realtime-Voice
MiniCPM-o 4.5
OpenBMB's 9B open model that sees, hears and talks at the same time — full-duplex voice and video conversation small enough to run on a Mac. Speech is English and Chinese only.
- Modality
- Realtime-Voice
- License
- Apache 2.0 (Open)
- Parameter size
- 9B
- Released
- February 3, 2026
- Last verified
- October 28, 2026
- Runs locally
- Yes
- Also handles
- vision, text
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 16 GB (24 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 12 GB (32 GB recommended) of video memory.
- No graphics card
- Not practical. Too slow without a graphics card or Apple Silicon.
Full voice conversation works on a Mac through the maker's llama.cpp-omni app (an M3 or newer with 16 GB for turn-by-turn, an M4 Max with 24 GB for live full-duplex), not through the plain Ollama chat.
Numbers are for: 4-bit GGUF via llama.cpp-omni for Mac and small cards; full-precision PyTorch for the 28 GB figure. Based on the maker’s own figures; where the maker gives only a graphics-card figure, the Mac memory is our estimate.
Strengths
- Apache 2.0 weights and code
- Full-duplex: listens to audio and video continuously while speaking, without blocking
- Voice cloning and role play from a short reference clip
- Runs on a Mac: half-duplex voice on an M3/M4/M5 with 16 GB RAM, full-duplex on an M4 Max with 24 GB
- Or on an NVIDIA card with 12 GB via llama.cpp-omni; Ollama, vLLM and SGLang supported
Weaknesses
- Spoken conversation in English and Chinese only
- Full-precision PyTorch inference needs an NVIDIA GPU with at least 28 GB
- Same team's MiniCPM-V 4.7 (2026-10-12) is newer, but it is a vision model without speech
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | Apache 2.0 |
| GitHub | local | Model guide, llama.cpp-omni and Mac Docker demo |
Official sources
- Model card model
- GitHub github
- Model guide docs
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: MiniCPM-o 4.5 Hugging Face model card (deployment section).
- — Initial entry. Added 2026-10-28 from the popularity sweep: 27k GitHub stars (the OpenBMB/MiniCPM-o repo now redirects to OpenBMB/MiniCPM-V, which hosts both lines), 280k Hugging Face downloads/month. Released 2026-02-03 per the README news at the time ('We open-source MiniCPM-o 4.5'); API access followed on 2026-05-17.
Esc