Qwen3-32B
- In Input modalities: Text Out Output modalities: Text
- Reasoning
- Reasoning can be off
- Tool use
- Structured output
- Open weights
Prices
3 offerings · prices as listed| Conditions/ 1M tokens | Input | Output |
|---|---|---|
| Standard no longer listed since Oct 11, 2026 | $0.16 | $0.64 |
Prices from qwen3-32b model (opens in a new tab) (checked Oct 8, 2026)
| Conditions/ 1M tokens | Input | Output |
|---|---|---|
| Standard no longer listed since Oct 11, 2026 | ¥2.00 | — |
| Reasoning off no longer listed since Oct 11, 2026 | — | ¥8.00 |
| Reasoning on no longer listed since Oct 11, 2026 | — | ¥20.00 |
Prices from 阿里云百炼：模型价格 (opens in a new tab) (checked Oct 6, 2026)
Standard: the price a request with no conditions pays (the headline price elsewhere on the site).
Details
- Knowledge cutoff
- Apr 2025
- Licence
- apache-2.0
- Family
- qwen
- Model ID
Not stated on Alibaba Qwen’s pages: max input, reasoning efforts, max thinking budget, built-in tools, aliases.
Run locally
Weights: Qwen/Qwen3-32B- Parameters 32.8B
- Dense
- Native precision BF16
- Layers 64
- KV cache per token 256 KB
- Native context 32K (model card)
- With YaRN 128K (model card)
- Config positions 40K (config.json max_position_embeddings)
- API context 128K (maker’s API)
- Needs
- ≈ 19.0 GB (weights 16.2 + KV cache 2.0 + overhead 0.8)
- Usable
- 21.6 GB (90% of its memory)
- Generation speed
- 26–39 tok/s
- Longest context
- 18K (longer doesn’t fit)
- First token
- about 1.06–1.59 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 61.0 GB | ≈ 66.1 | ✕Doesn’t fit | — |
| Q8_0maker’s file | 32.4 GB | ≈ 36.0 | ○Needs system memory 3–4 tok/s | — |
| FP8maker’s file | 32.0 GB | ≈ 35.6 | ○Needs system memory 3–4 tok/s | — |
| Q6_Kmaker’s file | 25.0 GB | ≈ 28.3 | ○Needs system memory 5–7 tok/s | — |
| Q5_K_Mmaker’s file | 21.6 GB | ≈ 24.7 | ○Needs system memory 8–13 tok/s | — |
| Q5_0maker’s file | 21.1 GB | ≈ 24.1 | ○Needs system memory 10–15 tok/s | — |
| Q4_K_Mmaker’s file | 18.4 GB | ≈ 21.3 | ●Runs well 23–35 tok/s | 9K |
| Q4_0 | 17.2 GB | ≈ 20.0 | ●Runs well 24–37 tok/s | 14K |
| INT4 (AWQ / GPTQ)maker’s file | 18.0 GB | ≈ 20.9 | ●Runs well 23–35 tok/s | 10K |
| IQ4_XS | 16.2 GB | ≈ 19.0 | ●Runs well 26–39 tok/s | 18K |
| Q3_K_Mquality drops | 14.9 GB | ≈ 17.7 | ●Runs well 28–42 tok/s | 23K |
| IQ3_XXSquality drops | 11.7 GB | ≈ 14.3 | ●Runs well 34–52 tok/s | 37K |
| Q2_Kquality drops | 11.4 GB | ≈ 14.0 | ●Runs well 35–52 tok/s | 38K |
| IQ2_XXSquality drops | 7.9 GB | ≈ 10.4 | ●Runs well 48–71 tok/s | 128K |
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
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 136.9
#102 of 118 listed models · CI 134.5–139.3
- LMArena Text 1,347
#126 of 163 listed models · CI 1,337–1,356 · 3,926 votes
- Epoch AI · GPQA Diamond 65.7%
#103 of 168 listed models · ±2.79
Also listed: reasoning off 54.1%
- Epoch AI · OTIS Mock AIME 2024–2025 66.9%
#94 of 159 listed models · ±6.04
Also listed: reasoning off 23.1%
- 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
- LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Oct 28, 2026
- EQ-Bench by Samuel J. Paech (eqbench.com; EQ-bench/EQ-bench-site on GitHub), MIT License. · License · as of Oct 28, 2026
- changes we made
Changes
All changes →- Context window 256K → 131.07K
- Max output 8.19K → 16.38K
Sources
- 阿里云百炼：模型价格in Chinese (opens in a new tab) help.aliyun.com · checked Oct 6, 2026
- Text generation models - Qwen, DeepSeek, GLM | Model Studio - Alibaba Cloud Model Studio - Alibaba Cloud Documentation Center (opens in a new tab) alibabacloud.com · checked Oct 10, 2026
- models.dev: qwen3-32b (community data, MIT) (opens in a new tab) github.com · checked Sep 30, 2026
- qwen3-32b - Alibaba Cloud Model Studio (opens in a new tab) docs.modelstudio.console.alibabacloud.com · checked Oct 9, 2026
- qwen3-32b model (opens in a new tab) alibabacloud.com · checked Oct 8, 2026
- Qwen/Qwen3-32B model card (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen3-32B config.json (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen3-32B files (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen3-32B-AWQ files (opens in a new tab) huggingface.co · checked Nov 3, 2026
- Qwen/Qwen3-32B-FP8 files (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen3-32B-GGUF files (opens in a new tab) huggingface.co · checked Nov 2, 2026
- Qwen/Qwen3-32B-MLX-4bit files (opens in a new tab) huggingface.co · checked Nov 3, 2026
- Qwen/Qwen3-32B-MLX-6bit files (opens in a new tab) huggingface.co · checked Nov 3, 2026
- Qwen/Qwen3-32B-MLX-8bit files (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen3-32B-MLX-bf16 files (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen3-32B model card (opens in a new tab) huggingface.co · checked Oct 31, 2026
- 模型上下架与更新-大模型服务平台百炼-阿里云_大模型服务平台百炼(Model Studio)-大模型服务平台百炼(Model Studio)-阿里云帮助中心in Chinese (opens in a new tab) help.aliyun.com · checked Oct 30, 2026
- Qwen/Qwen3-32B 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