What hardware do I need to run Qwen3.8-Max (open weights)?

At its most compressed (BF16 (as released)) it needs roughly 4800GB of VRAM. VRAM figures are approximate and depend on context length and settings.

Is Qwen3.8-Max (open weights) free for commercial use?

Commercial use is permitted, but with conditions. The weights are published under a custom Qwen3.8-Max licence rather than the Apache 2.0 that covers the rest of the Qwen line, including the Qwen3.8-27B. Read the licence before relying on it at scale.

Can I run Qwen3.8-Max (open weights) on Apple Silicon?

It can run on Apple Silicon through general runtimes, but it is not specifically optimised for it.

What is Qwen3.8-Max (open weights)'s context window?

Qwen3.8-Max (open weights) has a context window of 262,144 tokens, about 262k.

Qwen3.8-Max (open weights)

Name: Qwen3.8-Max (open weights)
Author: Alibaba

Alibaba · Text generation · MoE 2.4T · 262k context · Released 13 August 2026

Permitted with conditions Open weights

Qwen3.8-2.4T-A95B is the open-weight checkpoint of the hosted Qwen3.8-Max. It is a sparse mixture-of-experts with 512 experts, ten routed plus one shared active per token, a 262k native context that extends to around a million tokens, and it requires thinking mode for every interaction. Two things matter for anyone considering it. First, the open checkpoint is text-only: the vision that the hosted Max offers is not part of this release. Second, it was published in BF16 at roughly 4.8TB, which is multi-node cluster territory, not a machine you run at a desk. For nearly everyone, the runnable Qwen3.8-27B is the model to look at instead.

Strengths

A frontier-scale open-weight model that can be self-hosted with data kept in-house
Very long context, 262k native and extensible towards a million tokens
Sparse mixture-of-experts, so only 95B of the 2.4T parameters compute per token

Weaknesses

Firmly server-class, needing a multi-node cluster rather than a workstation card
The open checkpoint is text-only, unlike the multimodal hosted Max
Requires thinking mode for all interactions, adding latency and token use
A custom licence, a departure from the Apache 2.0 used elsewhere in the Qwen line

Hardware requirements

Quantisation	Approx. VRAM	Notes
BF16 (as released)	~4800GB	The released format, roughly 4.8TB of weights across many files. This is multi-node cluster territory. A 4-bit community quantisation would cut it substantially but was not part of the official release.

What you'd need to run this

Licence

Qwen3.8-Max License

Qwen3.8-Max (open weights): common questions

What hardware do I need to run Qwen3.8-Max (open weights)?: At its most compressed (BF16 (as released)) it needs roughly 4800GB of VRAM. VRAM figures are approximate and depend on context length and settings.
Is Qwen3.8-Max (open weights) free for commercial use?: Commercial use is permitted, but with conditions. The weights are published under a custom Qwen3.8-Max licence rather than the Apache 2.0 that covers the rest of the Qwen line, including the Qwen3.8-27B. Read the licence before relying on it at scale.
Can I run Qwen3.8-Max (open weights) on Apple Silicon?: It can run on Apple Silicon through general runtimes, but it is not specifically optimised for it.
What is Qwen3.8-Max (open weights)'s context window?: Qwen3.8-Max (open weights) has a context window of 262,144 tokens, about 262k.

Availability

Recommended for

Organisations with cluster-class infrastructure wanting a frontier-scale open model
Server-class self-hosting where data must stay in-house

Related models

Run it with

Related guides

Glossary

Our coverage

Catalogue entry last verified 14 August 2026. Specifications change; verify anything you are about to spend money on.