When are Qwen3.8-Max open weights released?

Alibaba launched Qwen3.8-Max via API and web on August 3, 2026 and said open weights would follow 'next week' — the week of August 10, 2026. A smaller Qwen3.8-27B is slated for open-weight release in the same window. As of August 6, 2026 the weights had not yet dropped; verify the exact date on qwen.ai before planning a self-host.

What license will Qwen3.8-Max use?

The license was still undisclosed at the August 3, 2026 launch. Prior Qwen releases used Apache 2.0 for most sizes, but Alibaba has not confirmed terms for the 2.4-trillion-parameter Max tier. Enterprise adopters should wait for the published license before assuming commercial-use rights.

Can I run Qwen3.8-Max locally?

Qwen3.8-Max is a 2.4-trillion-parameter MoE — self-hosting it needs a large multi-GPU cluster, so it is not a laptop model. The companion Qwen3.8-27B is the realistic local option: a dense-scale model aimed at single high-end GPUs and workstations, releasing open-weight the same week.

Quick Answer

Qwen3.8-Max Open Weights: Release Date & License

Published: August 6, 2026

The Short Answer

Alibaba released Qwen3.8-Max on August 3, 2026 through its API and chat, and promised open weights “next week” — the week of August 10, 2026. As of August 6, 2026, the weights are not yet public. A smaller Qwen3.8-27B is set for open-weight release in the same window and is the realistic local-run option.

What Shipped and When

Item	Status (as of Aug 6, 2026)
Qwen3.8-Max API / web	Live since Aug 3, 2026
Qwen3.8-Max open weights	Promised week of Aug 10, 2026
Qwen3.8-27B open weights	Promised week of Aug 10, 2026
License terms	Not yet disclosed
API pricing	Not confirmed at launch

What It Is

Qwen3.8-Max is a 2.4-trillion-parameter Mixture-of-Experts model — the largest open-weight model Alibaba has disclosed. It is multimodal (text, image, video input) and tuned for agentic computer use. Alibaba reports OSWorld-Verified 86.1, Terminal-Bench 2.1 86.6, PaperBench 93.0, and OmniDocBench 1.5 92.1. All of those are vendor numbers with no independent replication yet as of August 6, 2026.

Why the License Matters

For a model this large, the license decides whether companies can actually deploy it. Apache 2.0 (Qwen’s usual choice for smaller sizes) would make it a genuine open alternative to closed frontier models. A restrictive or usage-gated license would blunt that. Wait for the published terms before assuming commercial rights.

What to Do Now

Want to self-host a local model → target Qwen3.8-27B, not Max. The 2.4T MoE needs a multi-GPU cluster.
Evaluating for production → test via API today, but hold procurement decisions until the license and independent benchmarks land.
Tracking the drop → watch qwen.ai/research and the official Qwen accounts for the exact release date within the Aug 10 week.

Sources

MarkTechPost — Qwen3.8-Max release: marktechpost.com
SCMP — Qwen3.8-Max widely accessible ahead of open weights: scmp.com
Qwen — official research page: qwen.ai/research