Blog
Breaking Model NewsQwen 3.87 min readJuly 19, 2026

Qwen 3.8 Is Launching: Can You Run It Locally Yet?

Qwen announced a 2.4-trillion-parameter model and says open weights are coming soon. The announcement is official. A local download is not available yet.

Short answer: no, you cannot run Qwen 3.8 locally today. Qwen officially announced on July 19 that Qwen 3.8 is launching with 2.4T parameters and will go open-weight soon. As of this article, the official Qwen Hugging Face organization and GitHub do not publish a Qwen 3.8 checkpoint, model card, GGUF or MLX build. LocalClaw will not list it as a local model until those artifacts exist.

2.4T
Announced parameters
Soon
Open-weight status
No
Official GGUF / MLX
No
Local install today

What Qwen actually announced

On July 19, 2026, the official Qwen account posted that Qwen 3.8 is launching and going open-weight soon. The same announcement describes a 2.4T-parameter model and positions it against leading frontier systems.

That is the reliable core of the news. Qwen 3.8 is real, the parameter claim comes directly from Qwen, and an open-weight release is promised. It does not yet tell local users the architecture, active parameter count, context window, license, memory requirement or quantization path.

Qwen 3.8 local status: confirmed versus missing

QuestionStatus on July 19What it means
Is Qwen 3.8 officially announced?YesThe official Qwen account published the launch message.
Are official open weights downloadable?Not yetQwen says they are coming soon, not that they are already public.
Is there an official model card?Not yetArchitecture, active parameters, context and license remain unverified.
Is there an official GGUF or MLX?NoThere is no trustworthy local install target today.
Is it in the LocalClaw model list?NoLocalClaw requires a verifiable downloadable local artifact.

Why 2.4 trillion parameters does not answer the local question

Total parameters alone do not determine local hardware requirements. A dense 2.4T model would be far beyond consumer deployment. A highly sparse Mixture-of-Experts model could activate only a fraction of those weights per token and run much faster, but the full weight set still has to be stored and made available to the runtime.

Until Qwen publishes the architecture, we cannot responsibly estimate RAM, VRAM, disk size or tokens per second. Any specific local hardware table published before the model card would be speculation. Even extreme 1-bit compression of 2.4T raw weights would imply hundreds of gigabytes before runtime overhead, unless the released model uses a different structure or smaller variants.

Will there be a smaller Qwen 3.8 for laptops?

Qwen has not officially announced a 27B, 35B, 8B or other consumer-size Qwen 3.8 checkpoint in the launch post. The correct answer today is unknown. Qwen has historically released broad open model families, so smaller checkpoints are plausible, but LocalClaw will not turn that pattern into a claim.

What should you run locally instead?

Best current Qwen local alternatives

  • Qwen 3.6 27B: the strongest straightforward dense Qwen choice for 32GB-class machines.
  • Qwen 3.6 35B-A3B: efficient MoE behavior with only 3B active parameters and a practical local path.
  • Qwen 3 Coder 30B-A3B: a proven local coding and agentic option with broad GGUF support.
  • Bonsai 27B: a Qwen 3.6 27B-derived 1-bit and ternary release for unusually low-memory experiments.

What LocalClaw will verify next

Qwen 3.8 will enter the LocalClaw catalogue only after the official Qwen organization publishes downloadable weights with a clear model card and license. We will then verify the architecture, actual local size, context window, runtime support, GGUF or MLX availability and realistic RAM tiers.

This distinction matters for SEO and for users. A model announcement is news. A local model entry is a compatibility promise. Those should never be treated as the same thing.

LocalClaw verdict

Qwen 3.8 is one of the biggest AI model announcements of July 2026, and the promised open-weight release could become extremely important. But today it is a watchlist item, not a local recommendation. If a page claims you can already download Qwen 3.8 GGUF or run a verified 27B version, ask for the official Qwen repository.

For now, use Qwen 3.6 27B or 35B-A3B. They are real, documented and locally deployable. Return to this page when the Qwen 3.8 weights arrive: the hardware analysis will be updated from official artifacts, not rumors.

Sources