This model requires you to accept the Apache 2.0 license terms before download.
Description
Qwen3 30B-A3B Thinking — Alibaba Qwen Mixture-of-Experts LLM (30B total, 3B activated per token) tuned for deep step-by-step reasoning with explicit thinking traces, excelling at math, code and logic. Apache 2.0.
Tags
At a glance
| Kind | language model |
|---|---|
| Precision | bfloat16 |
| Container size | 56.9 GiB |
| Weights inside | 56.9 GiB |
| Components | 2 |
| Licence | Apache 2.0 |
What is inside the container
A .nbx container packs every part of Qwen3-30B-A3B-Thinking-2507 — 2 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| model | 0 | 48 | 56.3 GiB |
| lm_head | 0 | — | 594 MiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Max length
- 262144
- Temperature
- 0.6
- Top-p
- 0.95
- Top-k
- 20
Memory needed
The weights of Qwen3-30B-A3B-Thinking-2507 take 56.9 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 56.9 GiB (61,079,547,897 bytes).
$ sha256sum Qwen3-30B-A3B-Thinking-2507.nbx
0cc82fd592363f4b9883d636e114104d2b25753b3562be2edfc2771ab04c41e2
Questions
What is Qwen3-30B-A3B-Thinking-2507?
A language model by Qwen, distributed as a bfloat16 .nbx container and licensed Apache 2.0.
How big is the download?
56.9 GiB, of which 56.9 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/Qwen3-30B-A3B-Thinking-2507, then neurobrix serve --model Qwen3-30B-A3B-Thinking-2507.
CLI Usage
$ neurobrix import Qwen/Qwen3-30B-A3B-Thinking-2507
$ neurobrix serve --model Qwen3-30B-A3B-Thinking-2507
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop