Qwen3.6-35B-A3B
Alibaba · Qwen/Qwen3.6-35B-A3B
Weights license: Apache 2.0 · code and components may use other terms
Quality
No comparable score
We track no public, comparable benchmark that covers this model yet. No score is inferred.
Reported by the publisher
- SWE-bench Verified
- 73.4Qwen3.6-35BA3B · Language
- SWE-bench Multilingual
- 67.2Qwen3.6-35BA3B · Language
- SWE-bench Pro
- 49.5Qwen3.6-35BA3B · Language
- Terminal-Bench 2.0
- 51.5Qwen3.6-35BA3B · Language
- Claw-Eval Avg
- 68.7Qwen3.6-35BA3B · Language
- Claw-Eval Pass^3
- 50.0Qwen3.6-35BA3B · Language
- SkillsBench Avg5
- 28.7Qwen3.6-35BA3B · Language
- QwenClawBench
- 52.6Qwen3.6-35BA3B · Language
Every reported result
- NL2Repo
- 29.4Qwen3.6-35BA3B · Language
- QwenWebBench
- 1,397Qwen3.6-35BA3B · Language
- TAU3-Bench
- 67.2Qwen3.6-35BA3B · Language
- VITA-Bench
- 35.6Qwen3.6-35BA3B · Language
- DeepPlanning
- 25.9Qwen3.6-35BA3B · Language
- Tool Decathlon
- 26.9Qwen3.6-35BA3B · Language
- MCPMark
- 37.0Qwen3.6-35BA3B · Language
- MCP-Atlas
- 62.8Qwen3.6-35BA3B · Language
- WideSearch
- 60.1Qwen3.6-35BA3B · Language
- MMLU-Pro
- 85.2Qwen3.6-35BA3B · Language
- MMLU-Redux
- 93.3Qwen3.6-35BA3B · Language
- SuperGPQA
- 64.7Qwen3.6-35BA3B · Language
- C-Eval
- 90.0Qwen3.6-35BA3B · Language
- GPQA
- 86.0Qwen3.6-35BA3B · Language
- HLE
- 21.4Qwen3.6-35BA3B · Language
- LiveCodeBench v6
- 80.4Qwen3.6-35BA3B · Language
- HMMT Feb 25
- 90.7Qwen3.6-35BA3B · Language
- HMMT Nov 25
- 89.1Qwen3.6-35BA3B · Language
- HMMT Feb 26
- 83.6Qwen3.6-35BA3B · Language
- IMOAnswerBench
- 78.9Qwen3.6-35BA3B · Language
- AIME26
- 92.7Qwen3.6-35BA3B · Language
- MMMU
- 81.7Qwen3.6-35B-A3B · Vision Language
- MMMU-Pro
- 75.3Qwen3.6-35B-A3B · Vision Language
- Mathvista (mini)
- 86.4Qwen3.6-35B-A3B · Vision Language
- ZEROBench_sub
- 34.4Qwen3.6-35B-A3B · Vision Language
- RealWorldQA
- 85.3Qwen3.6-35B-A3B · Vision Language
- MMBench EN-DEV-v1.1
- 92.8Qwen3.6-35B-A3B · Vision Language
- SimpleVQA
- 58.9Qwen3.6-35B-A3B · Vision Language
- HallusionBench
- 69.8Qwen3.6-35B-A3B · Vision Language
- OmniDocBench1.5
- 89.9Qwen3.6-35B-A3B · Vision Language
- CharXiv (RQ)
- 78.0Qwen3.6-35B-A3B · Vision Language
- CC-OCR
- 81.9Qwen3.6-35B-A3B · Vision Language
- AI2D_TEST
- 92.7Qwen3.6-35B-A3B · Vision Language
- RefCOCO (avg)
- 92.0Qwen3.6-35B-A3B · Vision Language
- ODInW13
- 50.8Qwen3.6-35B-A3B · Vision Language
- EmbSpatialBench
- 84.3Qwen3.6-35B-A3B · Vision Language
- RefSpatialBench
- 64.3Qwen3.6-35B-A3B · Vision Language
- VideoMME (w sub.)
- 86.6Qwen3.6-35B-A3B · Vision Language
- VideoMME (w/o sub.)
- 82.5Qwen3.6-35B-A3B · Vision Language
- VideoMMMU
- 83.7Qwen3.6-35B-A3B · Vision Language
- MLVU
- 86.2Qwen3.6-35B-A3B · Vision Language
- MVBench
- 74.6Qwen3.6-35B-A3B · Vision Language
- LVBench
- 71.4Qwen3.6-35B-A3B · Vision Language
Values as printed on the card, with the publisher's own settings (card section: Benchmark Results › Language; Benchmark Results › Vision Language). Not comparable across publishers and not used in any ranking. Model card
Runs on
Fast memory holds it 33
The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.
Show all 33
Within memory, outside the fast part 50
Within the machine's memory, outside its fast part: on a graphics card the layers that do not fit run on the CPU from PC RAM (the listed RAM, or 32 GB assumed); on a Mac, above the macOS default GPU share. Not a speed figure.
Show all 50
One memory tier short 9
The same machines with the next memory size up would hold it in fast memory.
Show all 9
Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Not a measured fit or speed.
Alibaba released Qwen3.6-35B-A3B on 2026-04-16: a vision-language model that answers about images, with 36B parameters on the Hugging Face count and the Apache 2.0 license.
License: commercial use allowed. At Q4 the weights load about 27 GB before each machine's reserve; image resolution, video frames and context add memory, and the vision encoder needs support in the runtime you pick.
Measured speed
No published measurement for this checkpoint
Tokens per second are shown only when a source publishes them with hardware, build and setup. They are never estimated.
How to run
Runtime compatibility and tokens per second have not been measured for this entry. Published metadata and a memory estimate do not prove a successful local run.
Library in source: transformers
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
vllm serve Qwen/Qwen3.6-35B-A3B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 python -m sglang.launch_server --model-path Qwen/Qwen3.6-35B-A3B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3transformers serve Qwen/Qwen3.6-35B-A3B --port 8000 --continuous-batchingTechnical details
What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.
- Checkpoint / source
- Qwen/Qwen3.6-35B-A3B
- Task
- image-text-to-text
- Architecture
- Qwen3_5MoeForConditionalGeneration
- Library in source
- transformers
- Weight license
- Apache 2.0Qwen/Qwen3.6-35B-A3B model card (Hugging Face) · read 2026-10-18source
- Commercial use
- Allowed Apache License 2.0 text (SPDX) · read 2026-10-18license text
- Context
- 262,144 tokensNative window stated by the publishersource
- Active parameters (MoE)
- 3BQwen/Qwen3.6-35B-A3B model card (Hugging Face) · read 2026-10-09source
- Released
- QwenLM/Qwen3.5 README (GitHub) · read 2026-10-09source
Every stored field
- Revision checked
- 995ad96eacd98c81ed38be0c5b274b04031597b0
- Architecture
- Qwen3_5MoeForConditionalGeneration
- Library
- transformers
- Hub task tag
- image-text-to-text
- Formats
- transformers, safetensors
- Access
- Open download
- Publisher kind
- Official
- Hub parameter count
- 35,951,822,704safetensors.total
- Repository created
- 2026-04-15creation is not the model's release
- Repository updated
- 2026-04-24a commit is not a release date
- Hub downloads (30 days) · likes
- 3,971,758 · 3,029this repository only · read 2026-10-25
- Catalog state
- Hub config or weights found; no inventory benchmark link
Sources
Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.
- Publisher model card
- License: Qwen/Qwen3.6-35B-A3B model card (Hugging Face)
- Release date: QwenLM/Qwen3.5 README (GitHub)
Curated facts checked: 2026-10-18. The license and release date cite the publisher's sources; parameters, files and task come from the Hugging Face record, and memory is a labelled estimate from its parameter count. A value no source states stays empty; a repository commit is not a model release date.

























































