Qwen3.5-9B
Alibaba · Vision-language · 9B · 262k context · Released 1 March 2026
Qwen3.5-9B is the larger of the Qwen3.5 small dense models. It reads text, images and video and produces text, with a 262k context extensible towards a million tokens with YaRN, under an Apache 2.0 licence. Qwen's own figures put its general-knowledge score level with some older models several times its size, which is a real generational jump at this size. As with Qwen3.5-4B, it is the current Qwen dense model at 9B: later Qwen generations moved their small end to 27B dense and mixture-of-experts models rather than shipping a newer 9B.
Strengths
- Apache 2.0, freely usable commercially with no territory restriction
- Strong for its size, and the current Qwen dense model at 9B
- Fits a 16GB card, or 8GB at a 4-bit quantisation
- Multimodal input (image and video) with a long 262k context
Weaknesses
- Still a small model that hallucinates heavily on knowledge it was not trained on, so pair it with retrieval for factual work
- Vision support in local runtimes is less mature than text-only inference
- A 9B trails the larger open models on hard reasoning and agentic tasks
Hardware requirements
| Quantisation | Approx. VRAM | Notes |
|---|---|---|
| Q4_K_M | ~6GB | Fits an 8GB card with modest context. Approximate; the vision encoder adds a little |
| Q8_0 | ~10GB | Near-lossless, comfortable on a 16GB card. Approximate |
| BF16 | ~19GB | Full precision including the vision encoder, needs 24GB. Approximate |
Also runs on CPU (slower). Optimised builds available for Apple Silicon.
What you'd need to run this
Roughly what a machine to run this would need, at up to three levels of quality. Memory is the deciding factor.
Minimum to run it
Q4_K_M · ~6GB needed
One 12GB GPU
NVIDIA GeForce RTX 3060 12GBor a Mac or mini-PC with unified memory, if you prefer no discrete GPU, Mac mini M5 Pro .
At least 32GB of system RAM alongside the card.
around £700–£1,100
What else 12GB runs →For good quality
Q8_0 · ~10GB needed
One 16GB GPU
Intel Arc A770 16GBor a Mac or mini-PC with unified memory, if you prefer no discrete GPU, Mac mini M5 Pro .
At least 32GB of system RAM alongside the card.
around £700–£1,100
What else 16GB runs →Best quality
BF16 · ~19GB needed
One 24GB GPU
NVIDIA Tesla P40or a Mac or mini-PC with unified memory, if you prefer no discrete GPU, Mac mini M5 Pro .
32–64GB of system RAM alongside the card.
around £700–£1,100
What else 24GB runs →Licence
Apache 2.0 — read the licence
Benchmarks
| Benchmark | Score | Source | As of |
|---|---|---|---|
| MMLU-Pro | 82.5 | Qwen (model card) | March 2026 |
| GPQA Diamond | 81.7 | Qwen (model card) | March 2026 |
| LiveCodeBench v6 | 65.6 | Qwen (model card) | March 2026 |
| Artificial Analysis Intelligence Index | 32 | Artificial Analysis (independent) | March 2026 |
How it compares
How this model’s reported scores sit against other models we cover, on the same benchmarks. This model is highlighted.
LiveCodeBench v6
higher is better- Qwen3.8-27B 90.3
Qwen (model card) · August 2026
- Qwen3.5-9B 65.6
Qwen (model card) · March 2026
- Qwen3.5-4B 55.8
Qwen (model card) · March 2026
Artificial Analysis Intelligence Index
higher is better- Qwen3.5-9B 32
Artificial Analysis (independent) · March 2026
- Qwen3.5-4B 27
Artificial Analysis (independent) · March 2026
GPQA Diamond
higher is better- Qwen3.8-27B 89.2
Qwen (model card) · August 2026
- Muse Glimmer 30B 83.5%
Meta model card · August 2026
- Qwen3.5-9B 81.7
Qwen (model card) · March 2026
- gpt-oss-120b 80.1
OpenAI gpt-oss model card (high reasoning, no tools) · August 2025
- Qwen3.5-4B 76.2
Qwen (model card) · March 2026
- Nemotron 3.5 Lightning 75.44
NVIDIA (BF16) · August 2026
- DeepSeek-R1 71.5
DeepSeek-R1 technical report · January 2025
- gpt-oss-20b 71.5
OpenAI gpt-oss model card (high reasoning, no tools) · August 2025
- Qwen3 32B 68.4
Qwen3 technical report (thinking mode) · May 2025
- Qwen3 30B-A3B 65.8
Qwen3 technical report (thinking mode) · May 2025
- QwQ 32B 65.6
Qwen3 technical report (Table 13, QwQ-32B baseline) · May 2025
- Qwen3 14B 64.0
Qwen3 technical report (thinking mode) · May 2025
-
DeepSeek-R1 technical report (Table 5) · January 2025
- Qwen3 8B 62.0
Qwen3 technical report (thinking mode) · May 2025
- Gemma 3 27B 42.4
Gemma 3 technical report (27B IT) · March 2025
MMLU-Pro
higher is better- DeepSeek-R1 84.0
DeepSeek-R1 technical report · January 2025
- Qwen3.5-9B 82.5
Qwen (model card) · March 2026
- Nemotron 3.5 Lightning 81.94
NVIDIA (BF16) · August 2026
- Qwen3.5-4B 79.1
Qwen (model card) · March 2026
- QwQ 32B 69.07
Qwen model card · March 2025
- Gemma 3 27B 67.5
Gemma 3 technical report (27B IT) · March 2025
Qwen3.5-9B: common questions
- What hardware do I need to run Qwen3.5-9B?
- At its most compressed (Q4_K_M) it needs roughly 6GB of VRAM, and about 16GB for good quality. VRAM figures are approximate and depend on context length and settings.
- Is Qwen3.5-9B free for commercial use?
- Yes. Qwen3.5-9B is licensed under Apache 2.0, which permits commercial use with no meaningful conditions.
- Can I run Qwen3.5-9B on Apple Silicon?
- Yes. Qwen3.5-9B has builds optimised for Apple Silicon, through MLX or GGUF on a Mac.
- Does Qwen3.5-9B run on CPU?
- Yes, Qwen3.5-9B can run on the CPU, though generation is slower than on a GPU.
- What is Qwen3.5-9B's context window?
- Qwen3.5-9B has a context window of 262,144 tokens, about 262k.
Availability
Recommended for
- A capable general and multimodal model on a 16GB card
- Stronger small-model reasoning than the 4B, where you have the memory
- Long-context document and image work kept local
Related models
Related guides
Glossary
Catalogue entry last verified 12 October 2026. Specifications change; verify anything you are about to spend money on.