On this page
Qwen 3.5 runs on almost any Windows PC, offline and with no account. Off Grid AI Desktop runs every Qwen 3.5 from 0.8B to 27B for free. On 8 GB of RAM take the 4B, on 16 GB the 9B, and from 32 GB the 27B. After one download, it works with the internet off.
There’s no sign-up, no API key and no Python.
Download Off Grid AI Desktop (free, Windows x64)
Which Qwen 3.5 fits your PC?
Qwen 3.5 is Alibaba’s open model family from February 2026, under Apache 2.0 with a 262K context (9B model card). Every size reads images as well as text.
Off Grid AI Desktop judges fit by file size against your RAM. Up to 65% runs well, up to 82% is Tight, and past that it’s Too big.
| Model (in the app’s list) | Download | 8 GB PC | 16 GB PC | 32 GB PC |
|---|---|---|---|---|
| Qwen 3.5 0.8B | 0.7 GB | Runs well | Runs well | Runs well |
| Qwen 3.5 2B | 2.0 GB | Runs well | Runs well | Runs well |
| Qwen 3.5 4B | 3.8 GB | Runs well | Runs well | Runs well |
| Bonsai 2 9B (Qwen 3.5 9B, packed) | 3.0 GB | Runs well | Runs well | Runs well |
| Qwen 3.5 9B | 7.4 GB | Too big | Runs well | Runs well |
| Qwen 3.5 27B | 18.9 GB | Too big | Too big | Runs well |
Sizes are the full download from the app’s catalog, checked 16 November 2026. Each one is Unsloth’s 4-bit model file plus a vision file, and the 4B, 9B and 27B add a small speed-up file that earns them a Faster replies tag. The model files alone are 0.5, 1.3, 2.7, 5.7 and 16.7 GB. The app lists the 27B for 24 GB, where it’s Tight.
Bonsai 2 9B is Prism ML’s build of Qwen 3.5 9B with weights stored as -1, 0 or +1. It squeezes the 9B into 2.4 GB, or 3.0 GB with its vision file, and the app bundles the engine it needs on Windows. It’s the way to get a 9B onto an 8 GB laptop.
On a graphics card, the same sizes apply to VRAM. The 9B fits an 8 GB card and the 27B fits a 24 GB card.
How to set it up on Windows
- Download Off Grid AI Desktop and run the installer. There’s no account screen.
- Go through first-run setup. It picks a chat model for your RAM. It may choose Qwen 3.5 or another model, such as Bonsai 2 27B on a 16 GB PC.
- To pick the size yourself, open Models > Text and download the Qwen 3.5 size from the table.
- If you have an NVIDIA card, go to Settings → GPU performance, choose Download in Use your NVIDIA GPU, then Restart app. AMD and Intel GPUs use Vulkan, which is built in. Without a GPU, skip this.
- Start a new chat, pick the model and ask something.
Try one that uses the vision file. Take a screenshot with Win + Shift + S, attach it, and ask:
What does this error message mean, and what should I try first?
Qwen 3.5 thinks before it answers by default. For quicker replies, lower the Thinking budget in Model settings → Text.
Check that it really runs offline
- Make one successful chat with the model.
- Turn off Wi-Fi or unplug the network cable.
- Ask another question.
It answers, because the model and the engine are on your disk. Your chats stay on your PC, and nothing goes to Alibaba or to us.
What about the bigger Qwen 3.5 models?
Qwen 3.5 also comes as 35B-A3B, 122B-A10B and 397B-A17B. The 35B-A3B now has a newer version, Qwen 3.6 35B A3B, which is in the app’s list for 32 GB PCs. The 122B and 397B need far more memory than a Windows desktop has.
If you have 32 GB and want more than Qwen 3.5 27B, look at Qwen 3.8 27B, the newest model in the family, which the app also lists for 32 GB.
What goes wrong, and the fix
| Symptom | Check | Fix |
|---|---|---|
| Setup picked a different model | Your RAM and the performance mode | Download Qwen 3.5 from Models > Text yourself |
| Replies are slow | CPU only, or a size marked Tight | Drop one size, or add the NVIDIA download |
| Long pause before each answer | Qwen 3.5 thinks first | Lower the Thinking budget in Model settings → Text |
| Now shows CPU on an NVIDIA card | Whether the NVIDIA download finished | Download it in Settings → GPU performance and restart |
| Image questions fail | The vision file | Check it finished downloading with the model |
FAQ
Which Qwen 3.5 should I run on a Windows PC with 16 GB of RAM? Qwen 3.5 9B. Its 7.4 GB download runs well on 16 GB, reads images, and is the strongest Qwen 3.5 that fits.
Can Qwen 3.5 run on an 8 GB Windows laptop? Yes. Qwen 3.5 4B (3.8 GB with its vision file) runs well on 8 GB. The regular 9B is too big there, so use Bonsai 2 9B, a 2.4 GB build of Qwen 3.5 9B (3.0 GB with its vision file) in Off Grid AI Desktop’s list.
Do I need an account or internet to run Qwen 3.5? No account at all. You need internet once, to download the app and the model. After that it runs offline.
Do I need an NVIDIA GPU? No. The small Qwen 3.5 models run on the CPU. NVIDIA cards speed it up through an optional download, and AMD and Intel GPUs use Vulkan.
Can Qwen 3.5 read images on Windows? Yes. Every Qwen 3.5 size in the app comes with a vision file, so you can attach a screenshot or photo and ask about it.
Related guides
- How to Run Local AI on Your Windows PC in 2026
- How to Run Qwen Locally on Your Computer in 2026
- How to Run Qwen 3.8 27B on Windows in 2026
- How to Use Your NVIDIA GPU for Local AI on Windows in 2026
Download Off Grid AI Desktop, pick the Qwen 3.5 size from the table, and turn off Wi-Fi once it’s done.