This model requires you to accept the Apache 2.0 license terms before download.
Description
Qwen3-Omni 30B-A3B Instruct — quad-modal understanding (text, image, audio, video) + generative speech output (talker + MTP mono-step predictor + codec vocoder) on one build; thinker branch, MoE 128-expert
Tags
At a glance
| Kind | multimodal model |
|---|---|
| Precision | bfloat16 |
| Container size | 65.8 GiB |
| Weights inside | 65.7 GiB |
| Components | 12 |
| Licence | Apache 2.0 |
What is inside the container
A .nbx container packs every part of Qwen3-Omni-30B-A3B-Instruct — 12 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| code2wav | 0 | — | 412 MiB |
| thinker.audio_tower | 0 | — | 1.2 GiB |
| thinker.lm_head | 0 | — | 594 MiB |
| thinker.model | 0 | 48 | 56.3 GiB |
| thinker.visual | 0 | — | 1.0 GiB |
| talker.code_predictor | 0 | — | 270 MiB |
| talker.code_predictor.model.codec_embedding | 0 | — | 60.0 MiB |
| talker.codec_head | 0 | — | 6.0 MiB |
| talker.hidden_projection | 0 | — | 12.0 MiB |
| talker.model.codec_embedding | 0 | — | 6.0 MiB |
| talker.model | 0 | — | 5.9 GiB |
| talker.text_projection | 0 | — | 12.0 MiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Max length
- 65536
- Temperature
- 0.6
- Top-p
- 0.95
- Top-k
- 20
Memory needed
The weights of Qwen3-Omni-30B-A3B-Instruct take 65.7 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 65.8 GiB (70,604,596,479 bytes).
$ sha256sum Qwen3-Omni-30B-A3B-Instruct.nbx
d19b72ae15297495013d985d0bdd66a55b36051fb4c5b53a42e639b777e40448
Questions
What is Qwen3-Omni-30B-A3B-Instruct?
A multimodal model by Qwen, distributed as a bfloat16 .nbx container and licensed Apache 2.0.
How big is the download?
65.8 GiB, of which 65.7 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 Qwen/Qwen3-Omni-30B-A3B-Instruct, then neurobrix serve --model Qwen3-Omni-30B-A3B-Instruct.
CLI Usage
$ neurobrix import Qwen/Qwen3-Omni-30B-A3B-Instruct
$ neurobrix serve --model Qwen3-Omni-30B-A3B-Instruct
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop