Run Qwen 3.8 without a GPU.

Why these two are not laptop models

Qwen 3.8 A95B is 2.4 trillion parameters in total with roughly 95 billion active for any one token. The active count is what makes it quick; the total is what has to sit in memory the whole time, because the router can reach any of it for the next token. That is a rack, not a graphics card, and no quantization closes that gap.

There is a Qwen 3.8 that does fit on your own machine, and it is a different model: the dense 27B, Apache 2.0, which runs on a single card at 4-bit. If that is what you came for, read how to run Qwen 3.8 27B on your own computer instead. This page is about the two you cannot.

Max or A95B

They are not a big one and a small one. They are two flagships with different handling, and the thing that decides between them is usually the reasoning.

• Qwen 3.8 Max is the cheaper of the two per token and the one with a working off switch. It reasons by default, and the Think button really stops it, which on a model at this price is the largest lever you have on the bill.
• Qwen 3.8 A95B is the open-weight 2.4 trillion parameter checkpoint, 95 billion active, with a 262,144 token context. It reasons on every single turn and there is no way to stop it, so the Think button is locked on and you should budget for a reasoning pass in every answer.
• Neither reads images. Both are text in, text out. If you need to paste a screenshot, pick one of the vision models in the same picker.
• Both call tools natively, so Agent mode and the coding agent run on either one with nothing to configure.

The longer version, with what each of the two numbers in the names actually controls, is in Qwen 3.8 Max vs A95B.

What you get on any of them

• Both Qwen 3.8 flagships in the picker, next to the rest of the catalog: all 38 chat models, on a credit pack exactly as on the top plan.
• The image, video and audio studios on the same account and the same credit pool.
• A personal API key, which is one OpenAI-compatible endpoint for your editor, your scripts and any client that speaks the chat completions format.

How to start in three steps

1. Buy a €5 credit pack, or take the Hosted plan if you want a monthly budget. Card details go to Stripe and the payment is secured with 3D Secure.
2. The Studio opens on your account. Pick Qwen 3.8 Max or Qwen 3.8 A95B in the picker.
3. Type. On Max, turn the Think button off for anything that does not need working out, and watch the difference in the credit meter.

What it costs

The honest limits

Those token counts are a pack spent on one model and nothing else, which nobody does: input tokens draw from the same wallet, a long conversation carries its history into every turn, and one image render costs more than a long chat. A95B reasons on every turn and you cannot turn that off, so its real number is lower than the arithmetic suggests. Cloud runs on shared hardware, so at busy hours a request can wait a few seconds in a queue before it starts streaming. The catalog is curated: we host open-weight models we have priced and tested, so you pick from the list rather than uploading your own base model. Account data is hosted in the EU and you can delete it yourself. We never train on your data and we never sell it.

Questions