QwQ 32B Preview
Qwen · Apache 2.0
Reasoning-focused model that uses chain-of-thought natively. Builders pair this with a coding model for complex architecture decisions. Approaches o1-mini on reasoning benchmarks. Same size as Qwen 2.5 Coder 32B, so they compete for VRAM; run one at a time, not concurrently.
- Parameters
- 32.5B
- Architecture
- Dense
- Context
- 131,072 tokens
- Released
- 2024-11-28
- Engines
- llama.cpp, ollama, vLLM
- Builder Tools
- Open WebUI, LM Studio
Parameters
32.5B
VRAM
21.9 GB
Context
128K
Formats
5
GPUs
20
QwQ 32B Preview (32.5B) requires 21.9 GB VRAM at recommended quality (Q5_K_M). At efficient quality (Q4_K_M), it fits in 18.4 GB VRAM, making it compatible with the AMD Radeon RX 9060 XT 16GB. On NVIDIA Grace Blackwell Ultra GB300, expect approximately 140 tok/s at Q5_K_M. For the best experience, AMD AI Powerhouse ($1,818) is recommended.
Source: OwnRig methodology
21.9 GB
Q5_K_M
19.5 GB
128K tokens
Reasoning
VRAM Requirements
| Quality | Quantization | VRAM | File Size |
|---|---|---|---|
| full | Q8_0 | 34.6 GB | 32.5 GB |
| recommended | Q5_K_M | 21.9 GB | 19.5 GB |
| efficient | Q4_K_M | 18.4 GB | 16.3 GB |
| compressed | Q3_K_M | 14.8 GB | 12.7 GB |
| compressed | Q2_K | 11.6 GB | 9.8 GB |
Context Length Impact
KV cache VRAM at Q5_K_M quality. Longer context = more memory.
| Context | KV Cache | Total VRAM |
|---|---|---|
| 2K | 410 MB | 22.3 GB |
| 4K | 819 MB | 22.7 GB |
| 8K | 1.5 GB | 23.4 GB |
| 16K | 3.1 GB | 25 GBexceeds 24 GB |
| 32K | 6.1 GB | 28 GBexceeds 24 GB |
| 64K | 12.3 GB | 34.2 GBexceeds 24 GB |
| 128K | 24.6 GB | 46.5 GBexceeds 24 GB |
Compatible GPUs
20 devices| NVIDIA Grace Blackwell Ultra GB300 | Q5_K_M | 140 tok/s | Excellent |
| Apple M4 Max (128GB Unified) | Q8_0 | 14 tok/s | Good |
| Apple M4 Max (64GB Unified) | Q5_K_M | 17 tok/s | Good |
| Apple M4 Ultra (192GB) | Q8_0 | 21 tok/s | Good |
| NVIDIA GeForce RTX 4090 | Q4_K_M | 24 tok/s | Good |
| AMD Radeon RX 7900 XTX | Q4_K_M | 21 tok/s | Good |
| AMD Radeon Pro W7900 | Q5_K_M | 18 tok/s | Good |
| NVIDIA RTX PRO 6000 Blackwell | Q5_K_M | 50 tok/s | Good |
| NVIDIA RTX PRO 6000 Blackwell Max-Q | Q5_K_M | 46 tok/s | Good |
| AMD Radeon RX 9070 | Q2_K | – | Acceptable |
| AMD Radeon RX 9060 XT 16GB | Q2_K | – | Acceptable |
| AMD Radeon RX 7600 | Q2_K | 2 tok/s | Marginal |
| Apple M3 Pro (18GB Unified) | Q3_K_M | – | Not viable |
| NVIDIA GeForce RTX 3080 10GB | Q2_K | – | Not viable |
| NVIDIA GeForce RTX 4060 8GB | Q2_K | – | Not viable |
| NVIDIA RTX 4060 Laptop (40-60W) | Q2_K | – | Not viable |
| NVIDIA RTX 4070 Laptop (80-115W) | Q2_K | – | Not viable |
| NVIDIA GeForce RTX 4070 Ti 12GB | Q2_K | – | Not viable |
| AMD Radeon RX 9060 XT 8GB | Q2_K | – | Not viable |
| NVIDIA GeForce RTX 5060 8GB | Q2_K | – | Not viable |
Showing 20 of 20 entries
Builder Context
QwQ 32B Preview is commonly used with Open WebUI, LM Studio. For an AI coding workflow, pair it with an embedding model like nomic-embed-text for local RAG.
Recommended Builds
Complete PC builds that can run QwQ 32B Preview.
AI Builder Workstation
Run every AI tool you need. Nothing leaves your machine
Runs 10 models
Extreme AI Workstation
Dual GPUs that run the biggest AI models at a smart price
Runs 8 models
High-End AI Workstation
Chat, generate images, and code with AI, all at once
Runs 8 models
High-End Home AI Server
Your household's private AI: chatbots, code tools, and more
Runs 12 models
Mac Studio AI Builder
Plug in and run AI: silent, powerful, no assembly required
Runs 6 models
Mid-Range Home AI Server
Serve multiple AI models to every device at home
Runs 9 models
Frequently Asked Questions
- How much VRAM does QwQ 32B Preview need?
- QwQ 32B Preview requires 21.9 GB VRAM at recommended quality (Q5_K_M). At lower quality settings, it can fit in as little as 11.6 GB.
- What is the best GPU for QwQ 32B Preview?
- The NVIDIA Grace Blackwell Ultra GB300 delivers the best performance for QwQ 32B Preview, achieving 140 tok/s at Q5_K_M with an excellent rating.
- What quantization should I use for QwQ 32B Preview?
- For the best quality, use Q5_K_M (21.9 GB VRAM). If your GPU has limited VRAM, Q2_K (11.6 GB) is the most efficient option with acceptable quality.
- Is QwQ 32B Preview good for coding?
- Yes. QwQ 32B Preview is used with Open WebUI, LM Studio for local AI coding. For the best coding experience, pair it with an embedding model for local RAG.
Data confidence: estimated. Source
VRAM requirements are calculated from model parameters and may vary by inference engine, context length, and batch size. Performance estimates are based on community benchmarks and should be verified for your specific configuration.Qwen is a trademark of its respective owner. OwnRig is not affiliated with or endorsed by the model creator.