This model requires you to accept the Apache 2.0 license terms before download.
Description
MiniCPM-o 4.5 — omni build: tri-modal understanding (text, image, audio) plus generative voice output (speech codec AR + flow-matching token2wav vocoder, 24 kHz); SigLIP NaViT + perceiver resampler, whisper-medium audio, Qwen3-8B dense LM
Tags
At a glance
| Kind | multimodal model |
|---|---|
| Precision | bfloat16 |
| Container size | 19.3 GiB |
| Weights inside | 18.1 GiB |
| Components | 14 |
| Licence | Apache 2.0 |
| Origin | openbmb/MiniCPM-o-4_5 @ 1f761131fa83 |
What is inside the container
A .nbx container packs every part of MiniCPM-o-4_5 — 14 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| apm | 0 | — | 586 MiB |
| audio_projection_layer | 0 | — | 40.0 MiB |
| resampler | 0 | — | 170 MiB |
| vpm | 0 | — | 797 MiB |
| llm.lm_head | 0 | — | 1.2 GiB |
| llm.model | 0 | 36 | 14.1 GiB |
| tts.emb_code.0 | 0 | — | 9.6 MiB |
| tts.emb_text | 0 | — | 223 MiB |
| tts.head_code.0 | 0 | — | 9.6 MiB |
| tts.model | 0 | — | 407 MiB |
| tts.projector_semantic | 0 | — | 7.1 MiB |
| flow_encoder | 0 | — | 157 MiB |
| flow_dit | 0 | — | 437 MiB |
| hift | 0 | — | 79.5 MiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Max length
- 40960
- Temperature
- 0.6
- Top-p
- 0.95
- Top-k
- 20
Memory needed
The weights of MiniCPM-o-4_5 take 18.1 GiB in bfloat16. That is the floor for the memory the runtime has to place, spread over the devices it finds; activations come on top and grow with the size and length you ask for. A measured peak per device class is not published for this model yet.
Verify what you downloaded
Every download carries its SHA-256, and the file endpoint sends it in the X-NBX-SHA256 header. The container is 19.3 GiB (20,707,225,445 bytes).
$ sha256sum MiniCPM-o-4_5.nbx
a8e78addd92e429715dffdd82afecd0fb87467a5858fc302a2ee35bb9d3d6aba
Questions
What is MiniCPM-o-4_5?
A multimodal model by openbmb, distributed as a bfloat16 .nbx container and licensed Apache 2.0.
How big is the download?
19.3 GiB, of which 18.1 GiB are weights.
How do I run it?
Install the runtime from the download page, then use the commands under “CLI Usage” below: neurobrix import openbmb/MiniCPM-o-4_5, then neurobrix serve --model MiniCPM-o-4_5.
CLI Usage
$ neurobrix import openbmb/MiniCPM-o-4_5
$ neurobrix serve --model MiniCPM-o-4_5
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop