Qwen3.5-9B
- In Input modalities: TextImageVideo Out Output modalities: Text
- Reasoning
- Reasoning can be off
- Tool use
- Structured output
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Licence
- apache-2.0
- Family
- qwen
Not stated on Alibaba Qwen’s pages: max input, reasoning efforts, max thinking budget, built-in tools, knowledge cutoff, aliases.
Run locally
Weights: Qwen/Qwen3.5-9B- Parameters 9.7B
- Dense
- Native precision BF16
- Layers 32
- KV cache per token 32 KB
- Native context 256K (model card)
- Config positions 256K (config.json max_position_embeddings)
- API context 256K (maker’s API)
- Needs
- ≈ 19.1 GB (weights 18.0 (maker’s file) + KV cache 0.3 + overhead 0.9)
- Usable
- 21.6 GB (90% of its memory)
- Generation speed
- 26–39 tok/s
- Longest context
- 87K (longer doesn’t fit)
- First token
- about 0.31–0.47 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 18.0 GB | ≈ 19.1 | ●Runs well 26–39 tok/s | 87K |
| Q8_0 | 9.6 GB | ≈ 10.3 | ●Runs well 48–72 tok/s | 256K |
| FP8 | 9.0 GB | ≈ 9.7 | ●Runs well 51–76 tok/s | 256K |
| Q6_K | 7.4 GB | ≈ 8.1 | ●Runs well 61–92 tok/s | 256K |
| Q5_K_M | 6.4 GB | ≈ 7.1 | ●Runs well 71–106 tok/s | 256K |
| Q5_0 | 6.2 GB | ≈ 6.9 | ●Runs well 73–109 tok/s | 256K |
| Q4_K_M | 5.5 GB | ≈ 6.2 | ●Runs well 82–123 tok/s | 256K |
| Q4_0 | 5.1 GB | ≈ 5.8 | ●Runs well 88–133 tok/s | 256K |
| IQ4_XS | 4.8 GB | ≈ 5.5 | ●Runs well 93–140 tok/s | 256K |
| Q3_K_Mquality drops | 4.4 GB | ≈ 5.1 | ●Runs well 101–152 tok/s | 256K |
| IQ3_XXSquality drops | 3.4 GB | ≈ 4.2 | ●Runs well 127–191 tok/s | 256K |
| Q2_Kquality drops | 3.4 GB | ≈ 4.1 | ●Runs well 130–194 tok/s | 256K |
| IQ2_XXSquality drops | 2.3 GB | ≈ 3.1 | ●Runs well 183–274 tok/s | 256K |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): GPU GeForce RTX 3080; unified memory Apple M5 · 16 GB.
Run commands
vllm serve Qwen/Qwen3.5-9B --max-model-len 8192
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by us- Epoch Capabilities Index 139.3
#98 of 118 listed models · CI 137–141.5
- Epoch AI · GPQA Diamond 79%
#81 of 168 listed models · ±2.37
Also listed: reasoning off 78.9%
- Epoch AI · OTIS Mock AIME 2024–2025 61.7%
#102 of 159 listed models · reasoning off · ±6
Also listed: qwen3.5-9B 42.8%
- Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Nov 4 – 5, 2026
- changes we made
Changes
All changes →No changes recorded yet.
Sources
- Qwen3.5-9B (opens in a new tab) huggingface.co · checked Oct 30, 2026
- models.dev: alibaba/qwen3.5-9b (community data, MIT) (opens in a new tab) github.com · checked Oct 18, 2026
- Qwen/Qwen3.5-9B config.json (opens in a new tab) huggingface.co · checked Nov 3, 2026
- Qwen/Qwen3.5-9B model card (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen3.5-9B files (opens in a new tab) huggingface.co · checked Nov 3, 2026
- Qwen/Qwen3.5-9B README.md (opens in a new tab) huggingface.co · checked Nov 4, 2026
Alibaba Qwen’s official policies: data use, privacy, terms →
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
Alibaba Qwen · China · qwen.ai
Related models
Snapshots, the same family and other models from this provider