Qwen3.6 35B A3B — Hardware Requirements & GPU Compatibility
VisionFunctions
Qwen3.6 35B A3B pairs a mixture-of-experts design with roughly 35 billion total parameters, only about 3 billion of which are active per token. That small active footprint keeps generation speed close to a 3B-class dense model, even though the full parameter set must still be held in memory, similar to a dense model of that size, meaning local use still calls for a single high-end consumer GPU once quantized. The model also accepts image input alongside text, extending it to visual question answering and image-grounded chat. It supports a 256K token context window for long documents and extended conversations, and is released under the Apache 2.0 license. Published in April 2026, it gives Qwen 3.6 users a faster-inference alternative to the dense Qwen3.6 27B model released the same generation.
Runs well on a 24 GB GPU
Runs from 12 GB (IQ2_S · 32K)Best on 24 GB (Q4_K_M · ~151 tok/s · 128K)Macs from 32 GB (~24 tok/s)
Score 106.9#33 of 92 open#3 of 34 that fit 24 GBApache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
llmrun Composite Score →
106.9
- 90% range 103.9–110.0
- #33 of 92 open models
- #3 of 34 that fit 24 GB
Run Qwen3.6 35B A3B
Best fit for your MacBook Pro 14" M3 Max (96 GB). Every other method, OS and quant is in the step-by-step guide.
Recommended for your MacBook Pro 14" M3 Max (96 GB)
mlx_lm.chat --model lmstudio-community/Qwen3.6-35B-A3B-MLX-4bitMLX repo with its own quantisation
Download: Qwen3.6-35B-A3B-UD-Q4_K_M.gguf22.13 GBunsloth
Other ways
each opens step-by-stepRun in cloud
Fits on RTX PRO 4500 (32 GB) (10 GB headroom) · Q4_K_M
- Generation speed
- ~139 tok/s
- generation speed
- Cost per 1M output tokens
- $0.68
- per 1M output tokens
How Much VRAM Does Qwen3.6 35B A3B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On your device | |
|---|---|---|---|---|
| 15.7 GB | 15.9 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 256K context · ~56 tok/s | |
| 16.1 GB | 16.4 GB | 3-bit small — 3-bit small quantization | Fits: 256K context · ~55 tok/s | |
| 17.9 GB | 18.2 GB | 3-bit medium — 3-bit medium quantization | Fits: 256K context · ~52 tok/s | |
| 18.4 GB | 18.6 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 256K context · ~51 tok/s | |
| 21.9 GB | 22.2 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 256K context · ~46 tok/s | |
| 26 GB | 26.3 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 256K context · ~41 tok/s | |
| 30.0 GB | 30.3 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 256K context · ~37 tok/s | |
| 36.3 GB | 36.6 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 256K context · ~32 tok/s |
Selected Q4_K_M
4-bit medium quantization22.0GBat 16K context
- Weights
- 21.6 GB
- KV cache
- 0.2 GB
- Overhead
- 0.3 GB
Fits MacBook Pro 14" M3 Max (96 GB) · 70.5 GB free · ~46 tok/s
≥50 tok/s needs ≥250 GB/s · RTX 5090 ~213, RTX 3090 Ti ~151
Which Hardware Can Run Qwen3.6 35B A3B?
Q4_K_M · 21.9 GBQwen3.6 35B A3B (Q4_K_M) requires 21.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 29+ GB is recommended. Using the full 262K context window can add up to 2.6 GB, bringing total usage to 24.6 GB. 7 GPUs can run it, including NVIDIA GeForce RTX 5090. 103 devices with unified memory can run Qwen3.6 35B A3B, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Studio M5 Max (36 GB).
fits · 2.0 GB free · up to 128K ctx
6 GPUs
RTX 3090 Ti~151Simulate output on RTX 3090 TiRTX 4090~151Simulate output on RTX 4090RTX 3090~144Simulate output on RTX 309064–80 GB
fits · 42.0 GB free · up to 256K ctx
3 Mac classes · 1 device family
M2 Ultra*64 GB~76Simulate output on Mac Pro M2 Ultra (64 GB)M3 Max64 GB~55Simulate output on MacBook Pro 14" M3 Max (64 GB)Orin~42Simulate output on NVIDIA Jetson AGX Orin 64GB128–144 GB
fits · 106.0 GB free · up to 256K ctx
2 Mac classes · 4 device families
M2 Ultra*128 GB~76Simulate output on Mac Pro M2 Ultra (128 GB)M3 Max128 GB~55Simulate output on MacBook Pro 14" M3 Max (128 GB)GB10~54Simulate output on NVIDIA DGX SparkJetson T5000 (Thor)~54Simulate output on NVIDIA Jetson AGX Thor Developer Kit
By VRAM tier · typical GPUs and Macs at Q4_K_M
How good is Qwen3.6 35B A3B?
The llmrun Score merges 12 public benchmark boards into a single figure, where 100 represents the average for open models.
Score vs VRAM in the 24 GB tier
- Qwen3.6 35B A3B
- Other models
- Best Score for its VRAM
#3 of 34 that fit 24 GBEvery model here fits your MacBook Pro 14" M3 Max (96 GB) (96 GB)
Each dot is a model at the best quant that fits a 24 GB GPU. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.
Full 24 GB ranking →Benchmarks
The 6 boards that weigh most in the llmrun Score, of 12 with a result for this model; badges show who ran the test.
- GPQA Diamond
Epoch AI · Apr 2026Independent
Independent84.9%
#20/97 open
- AIME 2024/2025
Epoch AI · Apr 2026Independent
Independent86.7%
#19/85 open
- FrontierMath
Epoch AI · Apr 2026Independent
Independent20.3%
#16/16 open
- Terminal-Bench
Epoch AI · Apr 2026Lab-reported
Lab-reported23.0%
#13/16 open
- SciCode
Epoch AI · Apr 2026Lab-reported
Lab-reported35.8%
#41/65 open
- WeirdML
Epoch AI · Apr 2026Independent
Independent34.5%
#35/46 open
In the 24 GB class
34 models fit 24 GBSame Score, only models that fit 24 GB at their best quant: the ranking a 24 GB owner actually chooses from.
- #1
Qwen3.8 27BAlibaba
121.5
~38 tok/s · 17.9 GB
- #2
Qwen3.6 27BAlibaba
111.4
~38 tok/s · 17.9 GB
- #4
Gemma 4 31B ITGoogle
106.1
~32 tok/s · 21.7 GB
- #5
Qwen3.5 35B A3BAlibaba
105.6
~151 tok/s · 22.0 GB
- #6
Gemma 4 26B A4B ITGoogle
105.4
~171 tok/s · 16.5 GB
Alternatives to Qwen3.6 35B A3B
Qwen 3.6 family · 15 models →Official models with an llmrun Score, each answering one question; numbers are at the best quant that fits a 24 GB GPU.
Best alternative at this size
Qwen3.8 27B
Alibaba
- 14.6 higher Score
- 4.2 GB less VRAM
- 113 tok/s slower
Score 121.5 · 17.9 GB · ~38 tok/s
Fits in less VRAM
GPT OSS 20B
OpenAI
- 10.7 lower Score
- 8.9 GB less VRAM
- 39 tok/s faster
Score 96.2 · 13.1 GB · ~190 tok/s
Other vendor
Gemma 4 31B IT
Google
- 0.8 lower Score
- 0.3 GB less VRAM
- 119 tok/s slower
Score 106.1 · 21.7 GB · ~32 tok/s
Similar models
10| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen3.6 27BAlibaba | 111.4 | 17.9 GB | ~38 | similar score |
| Qwen3.5 35B A3BAlibaba | 105.6 | 22 GB | ~151 | similar score |
| Gemma 4 26B A4B ITGoogle | 105.4 | 16.5 GB | ~171 | similar score |
| Qwen3.5 27BAlibaba | 103.8 | 17.9 GB | ~38 | similar score |
| Seed OSS 36B InstructByteDance-Seed | 102.5 | 22.8 GB | ~31 | similar score |
| Qwen3 30B A3B Thinking 2507Alibaba | 100.6 | 19.4 GB | ~173 | similar score |
| Qwen3.5 9BAlibaba | 99.4 | 6.6 GB | ~103 | similar score |
| Qwen3 14BAlibaba | 97.9 | 11.9 GB | ~69 | similar score |
| Qwen3 8BAlibaba | 93.8 | 7.6 GB | ~119 | similar score |
| Qwen3 32BAlibaba | 98.4 | 22.6 GB | ~32 | same size |
Fine-tunes & variants(27)
- Aurora Code 1
- CyberStrike OffSec 35B
- Edge0 35B A3B Preview
- Grug 35B A3B
- Holo4 35B A3B
- Iris Mini
- Mihenk LLM v2 35B A3B Turkish Financial Model
- Qwen3.6 27B A3B Coder
- Qwen3.6 27B A3B CoderX
- Qwen3.6 28B
- Qwen3.6 28B REAP20 A3B
- Qwen3.6 35B A3B Abliterated V4
- Qwen3.6 35B A3B Claude 4.6 Opus Reasoning Distilled
- Qwen3.6 35B A3B Claude 4.7 Opus Reasoning Distilled
- Qwen3.6 35B A3B DFlash
- Qwen3.6 35B A3B Kimi K2.6 Reasoning Distilled
- Qwen3.6 35B A3B MicroTurn Duplex LoRA
- Qwen3.6 35B A3B REAP 90pct
- Qwen3.6 35B A3B StyleTune Voice
- Qwen3.6 35B A3B Uncensored Heretic Native MTP Preserved
- Qwen3.8 35B A3B Distill
- Qwen3.8 Distill 35B A3B Coder Abliterated
- SIQ 1 35B
- T Search
- Whittle Next 27B A3B
- WorldSim Opus 3.6 35B A3B
- Xor
Specifications
- Publisher
- Alibaba
- Family
- Qwen 3.6
- Parameters
- 36.0B
- Active parameters
- 4.7B per token
- Architecture
- Qwen3_5MoeForConditionalGeneration
- Context Length
- 262,144 tokens
- Vocabulary Size
- 248,320
- Release Date
- 2026-04-15
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Qwen3.6 35B A3B need?
Qwen3.6 35B A3B requires 21.9 GB of VRAM at Q4_K_M, or 72.3 GB at BF16. Full 262K context adds up to 2.6 GB (24.6 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 36.0B × 4.8 bits ÷ 8 = 21.6 GB
KV Cache + Overhead ≈ 0.3 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 2.9 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Qwen3.6 35B A3B?
Yes, at Q5_0 (22.9 GB) or lower. Higher quantizations like Q5_K_S (25.1 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Qwen3.6 35B A3B?
For Qwen3.6 35B A3B, Q4_K_M (21.9 GB) offers the best balance of quality and VRAM usage. Q4_K_L (22.4 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 10.3 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Qwen3.6 35B A3B on a Mac?
Yes, but only at lower quantizations. The smallest Mac that can run Qwen3.6 35B A3B is Mac Mini M1 (16 GB) at IQ2_XXS; 135 of the 156 Macs we list can run it at some quantization. For Q4_K_M (21.9 GB) you need a Mac with more unified memory.
Can I run Qwen3.6 35B A3B locally?
Yes — Qwen3.6 35B A3B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen3.6 35B A3B?
At Q4_K_M, Qwen3.6 35B A3B can reach ~116 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~151 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 2.9 GB active ÷ (8000 × 0.65) = 0.557 ms, plus 40 layers × 0.055 ms = 2.200 ms, so 1000 ÷ 2.757 ms = ~363 tok/s
Estimated speed at Q4_K_M (21.9 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen3.6 35B A3B?
At Q4_K_M, the download is about 21.57 GB. The full-precision BF16 version is 71.90 GB. The smallest option (IQ2_XXS) is 9.89 GB.
Which GPUs can run Qwen3.6 35B A3B?
7 consumer GPUs can run Qwen3.6 35B A3B at Q4_K_M. Top options include AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090.
Which devices can run Qwen3.6 35B A3B?
103 devices with unified memory can run Qwen3.6 35B A3B at Q4_K_M (21.9 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Asus ROG Flow Z13 (2025, Ryzen AI Max+ 395, 128 GB), Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.