Qwen3.5 9B — Hardware Requirements & GPU Compatibility
VisionFunctions
Qwen3.5-9B is Alibaba's dense 9-billion-parameter language model with a built-in vision encoder, part of the Qwen3.5 generation. This is the post-trained, instruction-tuned release, combining Gated DeltaNet and gated-attention layers for efficient long-context inference, and it can take images alongside text for tasks like visual question answering as well as general chat, coding, and tool use. It offers a 262,144-token native context window, extensible up to roughly 1,010,000 tokens, and is released under the Apache 2.0 license. At around 9.7 billion parameters, 4-bit quantization needs only about 5.6GB of memory, so it fits comfortably on a single mid-range consumer GPU.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 32K)Best on 24 GB (Q4_K_M · ~103 tok/s · 256K)Macs from 32 GB (~13 tok/s)
Score 99.4#49 of 92 open#10 of 34 that fit 24 GBApache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
llmrun Composite Score →
99.4
- 90% range 96.0–102.7
- #49 of 92 open models
- #10 of 34 that fit 24 GB
Run Qwen3.5 9B
Best fit for your MacBook Pro 16" M2 Pro (16 GB). Every other method, OS and quant is in the step-by-step guide.
Recommended for your MacBook Pro 16" M2 Pro (16 GB)
mlx_lm.chat --model lmstudio-community/Qwen3.5-9B-MLX-4bitMLX repo with its own quantisation
Download: Qwen3.5-9B-Q4_K_M.gguf5.68 GBunsloth
Other ways
each opens step-by-stepRun in cloud
Fits on RTX 3060 12GB (5 GB headroom) · Q4_K_M
- Generation speed
- ~37 tok/s
- generation speed
- Cost per 1M output tokens
- $0.52
- per 1M output tokens
How Much VRAM Does Qwen3.5 9B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On your device | |
|---|---|---|---|---|
| 4.7 GB | 5.5 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 128K context · ~30 tok/s | |
| 4.8 GB | 5.6 GB | 3-bit small — 3-bit small quantization | Fits: 128K context · ~29 tok/s | |
| 5.3 GB | 6.1 GB | 3-bit medium — 3-bit medium quantization | Fits: 128K context · ~27 tok/s | |
| 5.4 GB | 6.2 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 128K context · ~26 tok/s | |
| 6.4 GB | 7.2 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 128K context · ~22 tok/s | |
| 7.5 GB | 8.3 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 128K context · ~19 tok/s | |
| 8.5 GB | 9.3 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 64K context · ~16 tok/s | |
| 10.2 GB | 11 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 64K context · ~14 tok/s |
Selected Q4_K_M
4-bit medium quantization6.6GBat 16K context
- Weights
- 5.8 GB
- KV cache
- 0.5 GB
- Overhead
- 0.3 GB
Fits MacBook Pro 16" M2 Pro (16 GB) · 5.9 GB free · ~22 tok/s
≥50 tok/s needs ≥489 GB/s · RTX 5090 ~183, RTX 3090 Ti ~103
Which Hardware Can Run Qwen3.5 9B?
Q4_K_M · 6.4 GBQwen3.5 9B (Q4_K_M) requires 6.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 9+ GB is recommended. Using the full 262K context window can add up to 8.3 GB, bringing total usage to 14.7 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3070 Ti. 159 devices with unified memory can run Qwen3.5 9B, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, NVIDIA Jetson Orin Nano 8GB (Super).
≤ 11 GB
≤ 11 GB tier →fits · 1.4 GB free · up to 32K ctx
16 GPUs · 1 device family
RTX 3080~78Simulate output on RTX 3080RTX 2080 Ti~63Simulate output on RTX 2080 TiRTX 3070 Ti~62Simulate output on RTX 3070 TiOrin Nano~10Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)12 GB
12 GB tier →fits · 5.4 GB free · up to 128K ctx
10 GPUs · 1 device family
RTX 3080 Ti~93Simulate output on RTX 3080 TiRTX 5070~69Simulate output on RTX 5070RTX 4070~52Simulate output on RTX 4070iPhone~9Simulate output on Apple iPhone 17 Pro20 GB
fits · 13.4 GB free · up to 256K ctx
1 GPU
RX 7900 XT~76Simulate output on RX 7900 XT
By VRAM tier · typical GPUs and Macs at Q4_K_M
How good is Qwen3.5 9B?
The llmrun Score merges 9 public benchmark boards into a single figure, where 100 represents the average for open models.
Score vs VRAM in the 24 GB tier
- Qwen3.5 9B
- Other models
- Best Score for its VRAM
#10 of 34 that fit 24 GB12 of 34 fit your MacBook Pro 16" M2 Pro (16 GB) (16 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 9 with a result for this model; badges show who ran the test.
- GPQA Diamond
Epoch AI · Feb 2026Independent
Independent79.0%
#25/97 open
- AIME 2024/2025
Epoch AI · Feb 2026Independent
Independent61.7%
#36/85 open
- Terminal-Bench
Epoch AI · Feb 2026Lab-reported
Lab-reported9.2%
#15/16 open
- MMLU-Pro
TIGER-LabIndependent
Independent82.5%
#17/144 open
- SciCode
Epoch AI · Feb 2026Lab-reported
Lab-reported27.6%
#54/65 open
- CritPt
Epoch AI · Feb 2026Lab-reported
Lab-reported0.3%
#43/65 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
- #3
Qwen3.6 35B A3BAlibaba
106.9
~151 tok/s · 22.0 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 Qwen3.5 9B
Qwen 3.5 family · 34 models →Official models with an llmrun Score, each answering one question; numbers are at the best quant that fits a 24 GB GPU.
Runner-up at this size
Qwen3 8B
Alibaba
- 5.6 lower Score
- 1 GB more VRAM
- 16 tok/s faster
Score 93.8 · 7.6 GB · ~119 tok/s
Much faster, similar quality
Gemma 4 26B A4B IT
Google
- 6 higher Score
- 9.9 GB more VRAM
- 68 tok/s faster
Score 105.4 · 16.5 GB · ~171 tok/s
Other vendor
GPT OSS 20B
OpenAI
- 3.2 lower Score
- 6.5 GB more VRAM
- 87 tok/s faster
Score 96.2 · 13.1 GB · ~190 tok/s
Similar models
10| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen3 30B A3B Thinking 2507Alibaba | 100.6 | 19.4 GB | ~173 | similar score |
| Qwen3 14BAlibaba | 97.9 | 11.9 GB | ~69 | similar score |
| Qwen2.5 7B InstructAlibaba | 65.1 | 5.8 GB | ~131 | same size |
| Gemma 3 4B ITGoogle | 64.0 | 5 GB | ~231 | same size |
| Llama 3.1 8B InstructMeta | 62.5 | 7.9 GB | ~124 | same size |
| Meta Llama 3 8B InstructMeta | 61.7 | 7.9 GB | ~124 | same size |
| ZDTaichu5.0 9BTaichuAI | — | 9.2 GB | ~101 | same size |
| Ornith 1.5 9Bornith-ai | — | 6.6 GB | ~103 | same size |
| ZAYA1 VL 8BZyphra | — | 7.5 GB | ~346 | same size |
| Qwopus3.5 9B v3Jackrong | — | 6.6 GB | ~103 | same size |
Fine-tunes & variants(30)
- Balastmed 9B
- Carnice V1 9B Hermes Agent Stage2 Merged
- Claude OSS
- Clef Flash
- DQwen3.5 9B Base
- Hornybot Julian POV
- Hornybot Mara POV
- JackOD 9B Coder
- LensVLM 9B
- Mark 1x 9B
- MiMo V2.6 Distill Qwen 9B
- NeoHorse 1 9B
- OmniCoder 9B
- Opus Therapy 9B
- OxCoder 9B
- Qwable 9B Claude Fable 5
- Qwen3.5 9B Abliterated
- Qwen3.5 9B Claude 4.6 Opus Reasoning Distilled
- Qwen3.5 9B DFlash
- Qwen3.5 9B Distilled Coding Agentic
- Qwen3.5 9B Gemini 3.1 Pro Reasoning Distill
- Qwen3.5 9B Terminal Merge
- Qwen3.5 Medical Ft
- Qwen3.8 9B
- Qwen3.8 9B Distill
- Qwen3.8 9B Distill Uncensored Heretic
- Qwythos 9B Claude Mythos 5 1M
- Tema Q X2 Thinking
- Tmax 9B
- Zenith 9B CodeCore Merge
Specifications
- Publisher
- Alibaba
- Family
- Qwen 3.5
- Parameters
- 9.7B
- Architecture
- Qwen3_5ForConditionalGeneration
- Context Length
- 262,144 tokens
- Vocabulary Size
- 248,320
- Release Date
- 2026-02-27
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Qwen3.5 9B need?
Qwen3.5 9B requires 6.4 GB of VRAM at Q4_K_M, or 19.9 GB at BF16. Full 262K context adds up to 8.3 GB (14.7 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 9.7B × 4.8 bits ÷ 8 = 5.8 GB
KV Cache + Overhead ≈ 0.6 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 ≈ 8.9 GB (at full 262K context)
VRAM usage by quantization
What's the best quantization for Qwen3.5 9B?
For Qwen3.5 9B, Q4_K_M (6.4 GB) offers the best balance of quality and VRAM usage. Q4_K_L (6.5 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 3.2 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Qwen3.5 9B on a Mac?
Yes, but only at lower quantizations. The smallest Mac that can run Qwen3.5 9B is Mac Mini M1 (8 GB) at IQ2_XXS; 156 of the 156 Macs we list can run it at some quantization. For Q4_K_M (6.4 GB) you need a Mac with more unified memory.
Can I run Qwen3.5 9B locally?
Yes — Qwen3.5 9B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen3.5 9B?
At Q4_K_M, Qwen3.5 9B can reach ~755 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~103 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 6.36 × 0.65 = ~818 tok/s
Estimated speed at Q4_K_M (6.4 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen3.5 9B?
At Q4_K_M, the download is about 5.79 GB. The full-precision BF16 version is 19.31 GB. The smallest option (IQ2_XXS) is 2.65 GB.
Which GPUs can run Qwen3.5 9B?
52 consumer GPUs can run Qwen3.5 9B at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 7600. 40 GPUs have plenty of headroom for comfortable inference.
Which devices can run Qwen3.5 9B?
159 devices with unified memory can run Qwen3.5 9B at Q4_K_M (6.4 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Apple iPhone 17 Pro, Asus ROG Flow Z13 (2025, 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.