Qwen3-30B-A3B
Qwen · Apache 2.0
Mixture of Experts: 30B total parameters, 3B active per token.
Mixture-of-Experts architecture with ~3B active parameters per token and ~30B total; inference still loads the full expert pool for typical local stacks, so VRAM tracks total model size while compute per token stays efficient. High efficiency for its quality tier. Apache 2.0; 32K default and 128K max context.
- Parameters
- 30B
- Architecture
- MoE (3B active)
- Context
- 131,072 tokens
- Released
- 2025-04-29
- Engines
- llama.cpp, ollama, vLLM
- Builder Tools
- Continue, LM Studio, Open WebUI
Parameters
30B
VRAM
23 GB
Context
128K
Formats
4
GPUs
14
Qwen3-30B-A3B (30B) requires 23 GB VRAM at recommended quality (Q5_K_M). At efficient quality (Q4_K_M), it fits in 20 GB VRAM, making it compatible with the AMD Radeon RX 9060 XT 16GB. On NVIDIA RTX PRO 6000 Blackwell, expect approximately 278 tok/s at Q8_0. For the best experience, AMD AI Powerhouse ($1,818) is recommended.
Source: OwnRig methodology
23 GB
Q5_K_M
21.7 GB
128K tokens
Chat
VRAM Requirements
| Quality | Quantization | VRAM | File Size |
|---|---|---|---|
| full | Q8_0 | 34 GB | 32.5 GB |
| recommended | Q5_K_M | 23 GB | 21.7 GB |
| efficient | Q4_K_M | 20 GB | 18.6 GB |
| compressed | Q3_K_M | 16 GB | 14.7 GB |
Context Length Impact
KV cache VRAM at Q5_K_M quality. Longer context = more memory.
| Context | KV Cache | Total VRAM |
|---|---|---|
| 2K | 205 MB | 23.2 GB |
| 4K | 410 MB | 23.4 GB |
| 8K | 922 MB | 23.9 GB |
| 16K | 1.8 GB | 24.8 GBexceeds 24 GB |
| 32K | 3.5 GB | 26.5 GBexceeds 24 GB |
| 64K | 7 GB | 30 GBexceeds 24 GB |
| 128K | 14.1 GB | 37.1 GBexceeds 24 GB |
Compatible GPUs
14 devices| NVIDIA Grace Blackwell Ultra GB300 | Q8_0 | 145 tok/s | Excellent |
| NVIDIA RTX PRO 6000 Blackwell | Q8_0 | 278 tok/s | Excellent |
| NVIDIA RTX PRO 6000 Blackwell Max-Q | Q8_0 | 256 tok/s | Excellent |
| Apple M4 Ultra (192GB) | Q8_0 | 25 tok/s | Good |
| NVIDIA GeForce RTX 4090 | Q5_K_M | 25 tok/s | Good |
| AMD Radeon RX 7900 XTX | Q5_K_M | 22 tok/s | Good |
| Apple M4 Max (128GB Unified) | Q8_0 | 17 tok/s | Acceptable |
| Apple M4 Max (64GB Unified) | Q8_0 | 14 tok/s | Acceptable |
| AMD Radeon Pro W7900 | Q8_0 | 15 tok/s | Acceptable |
| Apple M3 Pro (18GB Unified) | Q4_K_M | 3 tok/s | Marginal |
| AMD Radeon RX 7600 | Q3_K_M | 2 tok/s | Marginal |
| AMD Radeon RX 9070 | Q3_K_M | – | Marginal |
| AMD Radeon RX 9060 XT 16GB | Q3_K_M | – | Marginal |
| AMD Radeon RX 9060 XT 8GB | Q3_K_M | – | Not viable |
Showing 14 of 14 entries
Builder Context
Qwen3-30B-A3B 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 Qwen3-30B-A3B need?
- Qwen3-30B-A3B requires 23 GB VRAM at recommended quality (Q5_K_M). At lower quality settings, it can fit in as little as 16 GB.
- What is the best GPU for Qwen3-30B-A3B?
- The NVIDIA RTX PRO 6000 Blackwell delivers the best performance for Qwen3-30B-A3B, achieving 278 tok/s at Q8_0 with an excellent rating.
- What quantization should I use for Qwen3-30B-A3B?
- For the best quality, use Q5_K_M (23 GB VRAM). If your GPU has limited VRAM, Q3_K_M (16 GB) is the most efficient option with acceptable quality.
- Is Qwen3-30B-A3B good for coding?
- Yes. Qwen3-30B-A3B 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.