Qwen3-32B Instruct
Qwen · Apache 2.0
Qwen 3 dense 32B instruct for high-quality local inference where VRAM allows; 32K default and 128K max context. Apache 2.0.
- Parameters
- 32B
- Architecture
- Dense
- Context
- 131,072 tokens
- Released
- 2025-04-29
- Engines
- llama.cpp, ollama, vLLM
- Builder Tools
- Continue, LM Studio, Open WebUI
Parameters
32B
VRAM
25 GB
Context
128K
Formats
4
GPUs
16
Qwen3-32B Instruct (32B) requires 25 GB VRAM at recommended quality (Q5_K_M). At efficient quality (Q4_K_M), it fits in 21.5 GB VRAM, making it compatible with the Apple M3 Pro (18GB Unified). On NVIDIA Grace Blackwell Ultra GB300, expect approximately 120 tok/s at Q8_0. For the best experience, High-End Home AI Server ($3,842) is recommended.
Source: OwnRig methodology
25 GB
Q5_K_M
23.2 GB
128K tokens
Chat
VRAM Requirements
| Quality | Quantization | VRAM | File Size |
|---|---|---|---|
| full | Q8_0 | 37 GB | 34.8 GB |
| recommended | Q5_K_M | 25 GB | 23.2 GB |
| efficient | Q4_K_M | 21.5 GB | 19.8 GB |
| compressed | Q3_K_M | 17.5 GB | 16 GB |
Context Length Impact
KV cache VRAM at Q5_K_M quality. Longer context = more memory.
| Context | KV Cache | Total VRAM |
|---|---|---|
| 2K | 205 MB | 25.2 GBexceeds 24 GB |
| 4K | 512 MB | 25.5 GBexceeds 24 GB |
| 8K | 1 GB | 26 GBexceeds 24 GB |
| 16K | 1.9 GB | 26.9 GBexceeds 24 GB |
| 32K | 3.8 GB | 28.8 GBexceeds 24 GB |
| 64K | 7.7 GB | 32.7 GBexceeds 24 GB |
| 128K | 15.4 GB | 40.4 GBexceeds 24 GB |
Compatible GPUs
15 devices| NVIDIA Grace Blackwell Ultra GB300 | Q8_0 | 120 tok/s | Excellent |
| NVIDIA GeForce RTX 5090 | Q4_K_M | 44 tok/s | Excellent |
| NVIDIA GeForce RTX 4090 | Q5_K_M | 25 tok/s | Good |
| NVIDIA GeForce RTX 4090 | Q4_K_M | 30 tok/s | Good |
| NVIDIA RTX PRO 6000 Blackwell | Q8_0 | 31 tok/s | Good |
| Apple M4 Max (128GB Unified) | Q8_0 | 14 tok/s | Acceptable |
| Apple M4 Max (64GB Unified) | Q8_0 | 14 tok/s | Acceptable |
| Apple M4 Ultra (192GB) | Q8_0 | 21 tok/s | Acceptable |
| AMD Radeon RX 7900 XTX | Q5_K_M | 17 tok/s | Acceptable |
| AMD Radeon Pro W7900 | Q8_0 | 12 tok/s | Acceptable |
| NVIDIA RTX PRO 6000 Blackwell Max-Q | Q8_0 | 29 tok/s | Acceptable |
| Apple M3 Pro (18GB Unified) | Q3_K_M | 2 tok/s | Marginal |
| AMD Radeon RX 7600 | Q3_K_M | 2 tok/s | Marginal |
| AMD Radeon RX 9070 | Q3_K_M | 4 tok/s | Marginal |
| AMD Radeon RX 9060 XT 16GB | Q3_K_M | 2 tok/s | Marginal |
| AMD Radeon RX 9060 XT 8GB | Q3_K_M | – | Not viable |
Showing 16 of 16 entries
Builder Context
Qwen3-32B Instruct is commonly used with Continue, LM Studio, Open WebUI. For an AI coding workflow, pair it with an embedding model like nomic-embed-text for local RAG.
Frequently Asked Questions
- How much VRAM does Qwen3-32B Instruct need?
- Qwen3-32B Instruct requires 25 GB VRAM at recommended quality (Q5_K_M). At lower quality settings, it can fit in as little as 17.5 GB.
- What is the best GPU for Qwen3-32B Instruct?
- The NVIDIA Grace Blackwell Ultra GB300 delivers the best performance for Qwen3-32B Instruct, achieving 120 tok/s at Q8_0 with an excellent rating.
- What quantization should I use for Qwen3-32B Instruct?
- For the best quality, use Q5_K_M (25 GB VRAM). If your GPU has limited VRAM, Q3_K_M (17.5 GB) is the most efficient option with acceptable quality.
- Is Qwen3-32B Instruct good for coding?
- Yes. Qwen3-32B Instruct is used with Continue, LM Studio, Open WebUI for local AI coding. For the best coding experience, pair it with an embedding model for local RAG.
Data confidence: community. 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.