Qwen2.5-VL-32B-Instruct
- Open weights
Prices
1 offering · prices as listedqwen2.5-vl-32b-instruct| Conditions/ 1M tokens | Input | Output |
|---|---|---|
| Standard | ¥8.00 | ¥24.00 |
Prices from Token-based deployment, LoRA deployment, pay-as-you-go deployment-Alibaba Cloud Model Studio(Model Studio)-阿里云帮助中心 (opens in a new tab) (checked Oct 2, 2026)
Standard: the price a request with no conditions pays (the headline price elsewhere on the site).
Details
- Reasoning
- —
- Tool use
- —
- Structured output
- —
- Licence
- apache-2.0
- Family
- qwen-vl
- Model ID
Not stated on Alibaba Qwen’s pages: max input, input modalities, output modalities, reasoning efforts, reasoning can be off, max thinking budget, built-in tools, knowledge cutoff, aliases.
Run locally
Weights: Qwen/Qwen2.5-VL-32B-Instruct- Parameters 33.5B
- Dense
- Native precision BF16
- Layers 64
- KV cache per token 256 KB
- Config positions 125K (config.json max_position_embeddings)
- API context 977K (maker’s API)
- Needs
- ≈ 19.4 GB (weights 16.6 + KV cache 2.0 + overhead 0.8)
- Usable
- 21.6 GB (90% of its memory)
- Generation speed
- 25–38 tok/s
- Longest context
- 17K (longer doesn’t fit)
- First token
- about 1.08–1.62 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 63.6 GB | ≈ 68.8 | ✕Doesn’t fit | — |
| Q8_0 | 33.1 GB | ≈ 36.8 | ○Needs system memory 2–4 tok/s | — |
| FP8 | 31.2 GB | ≈ 34.7 | ○Needs system memory 3–4 tok/s | — |
| Q6_K | 25.6 GB | ≈ 28.9 | ○Needs system memory 5–7 tok/s | — |
| Q5_K_M | 22.2 GB | ≈ 25.3 | ○Needs system memory 8–12 tok/s | — |
| Q5_0 | 21.4 GB | ≈ 24.5 | ○Needs system memory 9–13 tok/s | — |
| Q4_K_M | 18.9 GB | ≈ 21.9 | ○Needs system memory 21–32 tok/s | — |
| Q4_0 | 17.5 GB | ≈ 20.4 | ●Runs well 24–36 tok/s | 13K |
| INT4 (AWQ / GPTQ)maker’s file | 19.3 GB | ≈ 22.2 | ○Needs system memory 16–25 tok/s | — |
| IQ4_XS | 16.6 GB | ≈ 19.4 | ●Runs well 25–38 tok/s | 17K |
| Q3_K_Mquality drops | 15.2 GB | ≈ 18.0 | ●Runs well 27–41 tok/s | 22K |
| IQ3_XXSquality drops | 11.9 GB | ≈ 14.5 | ●Runs well 34–51 tok/s | 36K |
| Q2_Kquality drops | 11.7 GB | ≈ 14.3 | ●Runs well 34–51 tok/s | 37K |
| IQ2_XXSquality drops | 8.0 GB | ≈ 10.5 | ●Runs well 47–70 tok/s | 52K |
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 5090; unified memory Apple M5 Max (32-core GPU) · 36 GB.
Run commands
The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.
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- LMArena Vision 1,152
#73 of 89 listed models · CI 1,137–1,168 · 1,490 votes
- LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Oct 19, 2026
- changes we made
Changes
All changes →- New language model
Sources
- Text generation - Alibaba Cloud Model Studio (opens in a new tab) docs.modelstudio.console.alibabacloud.com · checked Oct 15, 2026
- Text generation models - Qwen, DeepSeek, GLM | Model Studio - Alibaba Cloud Model Studio - Alibaba Cloud Documentation Center (opens in a new tab) docs.modelstudio.console.alibabacloud.com · checked Oct 15, 2026
- Token-based deployment, LoRA deployment, pay-as-you-go deployment-Alibaba Cloud Model Studio(Model Studio)-阿里云帮助中心in Chinese (opens in a new tab) help.aliyun.com · checked Oct 2, 2026
- Qwen/Qwen2.5-VL-32B-Instruct model card (opens in a new tab) huggingface.co · checked Nov 1, 2026
- Qwen/Qwen2.5-VL-32B-Instruct config.json (opens in a new tab) huggingface.co · checked Nov 2, 2026
- Qwen/Qwen2.5-VL-32B-Instruct files (opens in a new tab) huggingface.co · checked Nov 1, 2026
- Qwen/Qwen2.5-VL-32B-Instruct-AWQ files (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen2.5-VL-32B-Instruct model card (opens in a new tab) huggingface.co · checked Nov 3, 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