AI Index

Qwen3-VL-32B-Instruct

Language Alibaba Qwen Verified Released Oct 21, 2025
  • In TextImageVideo Out Text
  • Tool use
  • Structured output
  • Open weights

Prices

3 offerings · prices as listed
Alibaba Cloud Model Studio Official International · Global · USD qwen3-vl-32b-instruct
Conditions/ 1M tokensInputOutput
Standard no longer listed since Oct 11, 2026 $0.16$0.64

Prices from Alibaba Cloud Model Studio pricing (checked Oct 11, 2026)

Alibaba Cloud Model Studio Official China · CNY qwen3-vl-32b-instruct
Conditions/ 1M tokensInputOutput
Standard no longer listed since Oct 11, 2026 ¥2.00¥8.00

Prices from 阿里云百炼：模型价格 (checked Oct 6, 2026)

Standard: the price a request with no conditions pays (the headline price elsewhere on the site).

Details

Family
qwen-vl
Model ID

Not stated on Alibaba Qwen’s pages: max input, reasoning, reasoning efforts, reasoning can be off, max thinking budget, built-in tools, knowledge cutoff, licence, aliases.

Run locally

Weights: Qwen/Qwen3-VL-32B-Instruct
  • Parameters 33.4B
  • Dense
  • Native precision BF16
  • Layers 64
  • KV cache per token 256 KB
  • Config positions 256K (config.json max_position_embeddings)
  • API context 128K (maker’s API)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Runs well IQ4_XS (automatic) 8K context 25–38 tok/sOne GPU
WeightsKV cacheOverheadUsable
Needs
≈ 19.3 GB (weights 16.5 + 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
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
BF16 (native)maker’s file62.1 GB≈ 67.2Doesn’t fit—
Q8_0maker’s file32.4 GB≈ 36.0Needs system memory 3–4 tok/s—
FP8maker’s file33.1 GB≈ 36.7Needs system memory 2–4 tok/s—
Q6_K25.5 GB≈ 28.8Needs system memory 5–7 tok/s—
Q5_K_M22.1 GB≈ 25.2Needs system memory 8–12 tok/s—
Q5_021.4 GB≈ 24.4Needs system memory 9–14 tok/s—
Q4_K_Mmaker’s file18.4 GB≈ 21.3Runs well 23–35 tok/s9K
Q4_017.5 GB≈ 20.3Runs well 24–36 tok/s13K
IQ4_XS16.5 GB≈ 19.3Runs well 25–38 tok/s17K
Q3_K_Mquality drops15.2 GB≈ 17.9Runs well 27–41 tok/s22K
IQ3_XXSquality drops11.9 GB≈ 14.5Runs well 34–51 tok/s36K
Q2_Kquality drops11.6 GB≈ 14.2Runs well 34–52 tok/s37K
IQ2_XXSquality drops8.0 GB≈ 10.5Runs well 47–70 tok/s52K

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 4090; unified memory Apple M6 (24GB or 32GB) · 32 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
Full≈ 500.3 GBDoesn’t fit your hardware
LoRA≈ 66.0 GBDoesn’t fit your hardware
QLoRA≈ 21.3 GBFits your hardware

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

This model isn’t on any leaderboard we cite yet.

Changes

All changes →

No changes recorded yet.

Sources

Alibaba Qwen’s official policies: data use, privacy, terms →

Alibaba Qwen · China · qwen.ai

Snapshots, the same family and other models from this provider