Alibaba Qwen · mixture of experts · Apache-2.0
What runs Qwen3 Coder Next
Coding-agent MoE with 80B total and 3B active parameters. Fits 64 GB Macs and GPU-plus-RAM builds at 4-bit.
- Parameters
- 80B, 3B active
- 4-bit file
- 45.2 GB
- Q4_K_M
- Needs at 8k context
- 46.9 GB
- context memory estimated
- Longest context
- 262,144 tokens
Context memory: 24,576 bytes per token for the full-attention layers only. Hybrid (Qwen3-Next layout): 12 of 48 layers are Gated Attention (2 x 12 x 2 KV heads x 256 x 2 = 24,576 bytes/token); 36 Gated DeltaNet layers hold a fixed-size state. Card gives no per-token figure. How context memory works.
Measured on this hardware
Machines that hold the 4-bit file with 8,192 tokens of context and have published runs on the same chip or cards, fastest first. Median of single-user runs with plain decoding.
| Machine | Fit | Price | Generation |
|---|---|---|---|
| Apple Mac Studio (M5 Ultra, 64-core GPU)Apple unified memory | Fitsassumed limitfrom 96 GB | ~$5,499–$9,499 | 111 tok/smedian of 1 run |
| Apple Mac Studio (M5 Ultra, 80-core GPU)Apple unified memory | Fitsassumed limitfrom 96 GB | ~$6,799–$10,799 | 102 tok/smedian of 2 runs |
| Apple Mac Studio (M5 Max, 40-core GPU)Apple unified memory | Fitsassumed limitfrom 64 GB | ~$3,099–$5,099 | 98 tok/smedian of 3 runs |
| Apple MacBook Pro 14-inch (M5 Max, 40-core GPU)Apple unified memory | Fitsassumed limitfrom 64 GB | ~$4,699–$6,699 | 98 tok/smedian of 3 runs |
| Apple MacBook Pro 16-inch (M5 Max, 40-core GPU)Apple unified memory | Fitsassumed limitfrom 64 GB | ~$4,999–$6,999 | 98 tok/smedian of 3 runs |
| Dell Pro Max Tower T2 with NVIDIA RTX PRO 6000 BlackwellNVIDIA workstation | Fits | ~$13,909 | 91 tok/smedian of 1 run |
| 1x RTX PRO 6000 BlackwellSingle GPU | Fits | ~$20,833 | 91 tok/smedian of 1 run |
| Apple Mac mini (M5 Pro, 20-core GPU)Apple unified memory | Fitsassumed limitfrom 64 GB | ~$1,899–$2,899 | 74 tok/smedian of 2 runs |
| NVIDIA DGX SparkNVIDIA GB10 | Fitsassumed limit | ~$4,699 | 49 tok/smedian of 2 runs |
| Dell Pro Precision with GB10NVIDIA GB10 | Fitsassumed limit | ~$5,523 | 49 tok/smedian of 2 runs |
Estimated, not measured
These fit but nobody has published a run we could match. The figure is memory bandwidth divided by the bytes read per token, scaled by the share of that ceiling measured runs reach. How estimates work.
| Machine | Fit | Price | Generation |
|---|---|---|---|
| Dell Pro Precision with GB300NVIDIA workstation | Fits | ~$159,283 | ~1,838–2,347 tok/sestimate |
| 2x RTX 5090Multi-GPU | Fits | ~$4,109 + unpriced parts | ~245–358 tok/sestimate |
| RTX 5090 + EPYC 9355P with 768 GB DDR5 (MoE offload)Hybrid GPU + RAM | GPU + RAM offloadassumed limit | ~$3,763 + unpriced parts | ~150–219 tok/sestimate, part in RAM |
| 4x Instinct MI50 32GB (used) on EPYC 7003Multi-GPU | Fits | ~$1,998 + unpriced parts | ~140–204 tok/sestimate |
| 4x RTX 3090 (used) on EPYC 7003, open frameMulti-GPU | Fits | ~$2,178 + unpriced parts | ~128–187 tok/sestimate |
| tiny corp tinybox red flashGPU desktop (prebuilt) | Fits | ~$22,000 | ~88–128 tok/sestimate |
| Framework Desktop (Ryzen AI Max+ PRO 495)AMD unified memory | Fitsassumed limit | ~$6,799 | ~71–90 tok/sestimate |
| Framework Desktop (Ryzen AI Max+ 395)AMD unified memory | Fitsassumed limitfrom 64 GB | ~$1,959–$3,449 | ~66–85 tok/sestimate |
| GMKtec EVO-X2 (Ryzen AI Max+ 395)AMD unified memory | Fitsassumed limitfrom 64 GB | ~$2,600–$4,000 | ~66–85 tok/sestimate |
| Beelink GTR9 Pro (Ryzen AI Max+ 395)AMD unified memory | Fits | ~$4,699 | ~66–85 tok/sestimate |
| Minisforum MS-S1 MAX (Ryzen AI Max+ 395)AMD unified memory | Fits | ~$4,899 | ~66–85 tok/sestimate |
| Corsair AI Workstation 300 (Ryzen AI Max+ 395)AMD unified memory | Fits | ~$4,800 | ~66–85 tok/sestimate |
| HP Z2 Mini G1a (Ryzen AI Max+ PRO 395)AMD unified memory | Fits | ~$4,999 | ~66–85 tok/sestimate |
| Apple Mac mini (M5 Pro, 16-core GPU)Apple unified memory | Fitsassumed limitfrom 64 GB | ~$1,699–$2,699 | ~56–100 tok/sestimate |
| RTX 3090 (used) + 128 GB DDR5 on AM5 (MoE offload)Hybrid GPU + RAM | GPU + RAM offloadassumed limit | ~$4,833 + unpriced parts | ~25–36 tok/sestimate, part in RAM |
Fits, no speed figure
Too large for
Apple Mac Studio (M5 Max, 32-core GPU), Apple MacBook Pro 14-inch (M5 Max, 32-core GPU), Apple MacBook Pro 16-inch (M5 Max, 32-core GPU), Framework Desktop (Ryzen AI Max 385), 1x RTX 3090 (used), 2x RTX 3090 (used), 1x RTX 5090, 2x Radeon RX 7900 XTX, 2x Intel Arc Pro B60. A smaller quant may fit: try the calculator.
Files
| Quant | Format | Size | Repo |
|---|---|---|---|
| 4bit | MLX | 41.8 GB | mlx-community/Qwen3-Coder-Next-4bit |
| Q4_K_M | GGUF | 45.2 GB | unsloth/Qwen3-Coder-Next-GGUF |
| Q5_K_M | GGUF | 52.9 GB | unsloth/Qwen3-Coder-Next-GGUF |
| Q6_K | GGUF | 61.1 GB | unsloth/Qwen3-Coder-Next-GGUF |
| 8bit | MLX | 78.8 GB | mlx-community/Qwen3-Coder-Next-8bit |
| Q8_0 | GGUF | 79.0 GB | unsloth/Qwen3-Coder-Next-GGUF |
Sources
- huggingface.co: model card: parameters, context length, license and attention layout; release month
- huggingface.co: downloads (last 30 days), license tag, safetensors param count
- huggingface.co: layer count, KV heads and head size
- huggingface.co: GGUF file sizes (shards summed)
- huggingface.co: MLX 4bit safetensors sizes
- huggingface.co: MLX 8bit safetensors sizes