Qwen 3.6 35B-A3B
Qwen 3.6 35B-A3B is a mixture-of-experts model: it has 35 billion parameters but computes each request with only three billion. That makes it answer much faster than dense models of similar size and suits many requests in a row. It still needs memory for all parameters, about 26 GB when quantised.
Profile
- Maker
- Alibaba (Qwen) · China
- Licence
- Free for commercial use
- Runs on
- Workstation with 64 GB
- Memory at 4 bits
- 26 GB
- Memory at 8 bits
- 37 GB
- Parameters
- 36bn, 3bn active
- Context
- 262,000 tokens
- Inputs
- text, image
- Released
- April 2026
Scores
- Epoch index
- 144.1 · rank 26 of 127 open
- GPQA Diamond (expert questions)
- 79.8 %
- AIME (maths)
- 86.7 %
Scores: LMArena, Epoch AI, EuroEval, Hugging Face, models.dev; as of 19 October 2026. Sources and method →
Use cases
Where the model makes sense.
Sorting and extracting at scale
- Situation
- A company has thousands of emails, tickets or reviews to sort by topic, urgency or sentiment.
- What it looks like
- A script sends each text to the model and gets back a category and a few fields as JSON. Thanks to the few active parameters, this runs quickly even on a workstation.
- What to watch
- Check a sample by hand to see whether the categories fit before results feed into reports.
Chat assistant for a whole team
- Situation
- Ten to twenty staff should use an internal AI chat at the same time.
- What it looks like
- The model runs on a workstation or Mac, and an interface such as LibreChat or Open WebUI handles login and history.
- What to watch
- With many concurrent users, keep an eye on response times and move to a server with a GPU early.
Product data from images
- Situation
- For a catalogue or shop, colour, material and features should be captured from many product photos.
- What it looks like
- The model describes each photo to a fixed schema; results go into the inventory system as drafts.
- What to watch
- Models do not reliably detect fine differences in colours and materials on photos.
Batch translation and summaries
- Situation
- Reports, minutes or manuals should be summarised or translated regularly.
- What it looks like
- A nightly run processes new documents and stores summaries or translations in a folder.
- What to watch
- Have a person proofread translations for customers or legally relevant texts.
The use cases were written with AI assistance (Claude by Anthropic) and reviewed by vona. Scores come unchanged from the sources listed.
When another model fits better
For demanding analysis and coding, the dense Qwen 3.8 27B is more accurate but takes more compute per answer.
How to start
ollama run qwen3.6:35bWith Ollama it fits tightly on a computer with 32 GB; with 64 GB there is room for longer contexts and several users.
Suitable interfaces and programs from our directory: Ollama, LibreChat, Open WebUI
Licence: Free for commercial use
The licence allows commercial use, modification and redistribution; the main obligation is to keep the licence notice.
Hugging Face ↗Qwen as a cloud service in the AI tools directory →Similar models
Comparable models.
Alibaba (Qwen) · China
Qwen 3.8 27B
Compact model from the Qwen 3.8 series that understands images and calls tools. Quantised, it runs on a computer with 32 GB and reaches scores otherwise seen from much larger models.
Free for commercial use · Runs on: Computer with 32 GB
Use cases and details →Google · United States
Gemma 4 26B-A4B
Mixture-of-experts variant of Gemma 4 with four billion active parameters. It responds faster than Gemma 4 31B with similar memory needs.
Free for commercial use · Runs on: Computer with 32 GB
In the overview →NVIDIA · United States
Nemotron 3 Nano 30B-A3B
Small NVIDIA mixture-of-experts text model with three billion active parameters, built for fast agents and tool calls on your own hardware.
Commercial use with conditions · Runs on: Workstation with 64 GB
In the overview →
FAQ
What hardware do I need for Qwen 3.6 35B-A3B?
In the common 4-bit version, Qwen 3.6 35B-A3B needs about 26 GB of memory including headroom for context and runtime. That fits: Workstation with 64 GB. The more accurate 8-bit version needs about 37 GB. Long documents increase the requirement.
May I use Qwen 3.6 35B-A3B commercially?
The licence allows commercial use, modification and redistribution; the main obligation is to keep the licence notice.
How good is Qwen 3.6 35B-A3B in German?
There are no independent German scores for Qwen 3.6 35B-A3B yet. Test it with your own examples before using it.
AI consulting
Your own language model for your company?
Together we find the model that fits your tasks, your hardware and your data protection requirements, test it with real documents and set up operation.