Qwen3-Omni-30B-A3B
Qwen3-Omni-30B-A3B is an open-weights language model from Qwen, released September 22, 2025. It has 35B parameters (mixture-of-experts, 3B active). The context window is 64K tokens.
We present Qwen3-Omni, a single multimodal model that, for the first time, maintains state-of-the-art performance across text, image, audio, and video without any degradation relative to single-modal counterparts.
Check it against your own PC →
Can you run Qwen3-Omni-30B-A3B locally?
The smallest common GPU it runs on is 6 GB of VRAM (with 32 GB of RAM): IQ3_XXS, GPU + experts in RAM, roughly 48 tokens per second. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Runs · GPU + CPU experts | IQ3_XXS (noticeable loss) | 5.2 GB | 10.5 GB | ~48 |
| 8 GB VRAM | Runs · GPU + CPU experts | Q5_K_M (excellent) | 7.2 GB | 20.1 GB | ~30 |
| 12 GB VRAM | Runs · GPU + CPU experts | Q6_K (≈ lossless) | 11.2 GB | 19.9 GB | ~31 |
| 16 GB VRAM | Runs · GPU + CPU experts | Q6_K (≈ lossless) | 15.2 GB | 15.9 GB | ~38 |
| 24 GB VRAM | Runs on GPU | IQ4_XS (good) | 20.9 GB | – | ~260 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 31.1 GB | – | ~300 |
Ready-made quantizations on Hugging Face: ggml-org/Qwen3-Omni-30B-A3B-Instruct-GGUF · steampunque/Qwen3-Omni-30B-A3B-Instruct-MP-GGUF · TrevorJS/Qwen3-Omni-30B-A3B-GGUF · cudabenchmarktest/Qwen3.8-27B-E03-Obliterated-Omni-Audio-Bridge-GGUF · MIThridate/Qwen3-Omni-35B-A3B-SAMI-AsymQ · Zero-000/OmniLlama-Qwen3-Omni-30B-A3B-GGUF
Benchmark results
Epoch AI has not published results for Qwen3-Omni-30B-A3B yet. New models are usually evaluated within days to weeks.
Links
Announcement · Hugging Face: Qwen/Qwen3-Omni-30B-A3B-Instruct
More from Qwen
Qwen3.8 Omni Flash · Qwen3.8 Max (0902) · Qwen3.8 Flash · Qwen3.8 Flash Next · Qwen 3.8 27B · Qwen3.8-2.4T-A95B · Qwen 3.8 Max · Qwen3.7 Flash