open-weightopen-weight · Qwen3.5 0.8B–9B · 4B is Apache 2.0
Qwen
the local one
This quiz’s pick for anyone who needs AI to run on their own device, with no external AI service. Qwen is a family, not one download. Small Qwen3.5 checkpoints have documented local setups, but they won’t match a flagship hosted model.
the short version
best forpeople who must keep everything on their own device or approved infrastructure.
skip it ifyou need flagship-level answers, or your device has very little memory.
who it’s best for
1
The privacy-first worker
Data can’t leave the machine. Run it locally with remote tools and web search turned off.2
The offline learner
Download once while online, then chat and work with local documents offline.3
The tinkerer with a decent laptop
Start with a 2B or 4B model in LM Studio or Ollama and see what your machine can handle.what it’s good at
Runs on your own device, no external AI service
Works offline after setup
Small checkpoints from 1 GB to 6.6 GB
No gaming GPU needed. CPU and regular memory work.
watch out for
A small local model won’t match a flagship hosted model.
SSD space or swap isn’t the same as memory.
Local isn’t automatic approval. Check device security, logs and backups.
what can your device run?
A practical start is a computer with 8 GB RAM and about 10 GB free. 16 GB gives more headroom. These are planning estimates for short chats, not Qwen-certified minimums or speed guarantees.
| Qwen3.5 | download | trial floor | prefer | free disk |
|---|---|---|---|---|
| 0.8B | 1.0 GB | 4 GB (experimental) | 8 GB | 10 GB |
| 2B | 2.7 GB | 8 GB | 8–16 GB | 10 GB |
| 4B | 3.4 GB | 8 GB, tight | 16 GB | 10 GB |
| 9B | 6.6 GB | 16 GB | 24–32 GB | 15 GB |
Downloads are Ollama catalog sizes. Memory and disk are planning estimates. 4 GB with 0.8B is experimental only.
Apple Silicon Mac8 GB: try 0.8B or 2B. 16 GB: start with 4B. LM Studio or Ollama.
Intel MacTry 0.8B or 2B with Ollama (CPU only). LM Studio doesn’t support Intel Macs.
Windows PC8 GB: small-model trial. 16 GB: prefer 4B. LM Studio x64 needs AVX2.
Linux or phoneLinux: match your architecture and RAM. Phones: PocketPal, but there’s no verified minimum. Test first.
how to get it
freeFree weights. Install LM Studio or Ollama, or PocketPal on a phone.
paidno subscription. You pay in hardware and power.
API per 1M tokensnot used here. This quiz runs Qwen locally.
whereLM Studio, Ollama (ollama run qwen3.5:2b), PocketPal
the quiz picks it first for
any + C · any task, when it must run on your own device