unsloth
Qwen3 Coder Next
Qwen3 Coder Next is a 79.7B mixture-of-experts qwen3next model from unsloth, with about 13.5B parameters active per token. Configured context length is 262,144 tokens. License: apache-2.0. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters79.7B / 13.5B active
Context256K
Attentiongqa
Licenseapache-2.0
48 layers · 2 KV heads · head dim 256 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 148.5 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 80.4 GB | Hugging Face file bytes |
| Q8_0 | 79.0 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 68.1 GB | Hugging Face file bytes |
| UD-Q6_K | 61.3 GB | Hugging Face file bytes |
| Q6_K | 61.1 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 55.5 GB | Hugging Face file bytes |
| UD-Q5_K_M | 55.2 GB | Hugging Face file bytes |
| Q5_K_M | 52.9 GB | Hugging Face file bytes |
| UD-Q5_K_S | 52.0 GB | Hugging Face file bytes |
| Q5_K_S | 51.3 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 46.2 GB | Hugging Face file bytes |
| UD-Q4_K_M | 45.9 GB | Hugging Face file bytes |
| Q4_K_M | 45.2 GB | Hugging Face file bytes |
| UD-Q4_K_S | 42.9 GB | Hugging Face file bytes |
| Q4_K_S | 42.4 GB | Hugging Face file bytes |
| UD-IQ4_XS | 35.8 GB | Hugging Face file bytes |
| IQ4_NL | 78.6 GB | Hugging Face file bytes |
| IQ4_XS | 39.7 GB | Hugging Face file bytes |
| Q4_1 | 46.6 GB | Hugging Face file bytes |
| Q4_0 | 42.2 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 33.8 GB | Hugging Face file bytes |
| Q3_K_M | 69.2 GB | Hugging Face file bytes |
| UD-IQ3_S | 27.7 GB | Hugging Face file bytes |
| Q3_K_S | 63.3 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 26.5 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 24.9 GB | Hugging Face file bytes |
| Q2_K | 54.5 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 21.7 GB | Hugging Face file bytes |
| UD-IQ1_M | 20.2 GB | Hugging Face file bytes |
| UD-IQ1_S | 20.0 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Partial offload | UD-IQ3_S | 11–15 |
| GeForce RTX 3060 12GB | Expert offload | UD-IQ3_S | 34–46 |
| GeForce RTX 4060 Ti 16GB | Expert offload | UD-IQ4_XS | 22–30 |
| GeForce RTX 4090 | Tight fit | UD-IQ1_M | 124–169 |
| GeForce RTX 5090 | Tight fit | UD-IQ3_S | 169–230 |
| Apple M4 Pro | Won't fit | UD-IQ1_S | — |
| Apple M5 | Won't fit | UD-IQ1_S | — |
| CPU only · DDR5-5600 | Runs on CPU | UD-IQ2_XXS | 10–14 |
Smallest devices that fit it fully
- GeForce RTX 4090 — Tight fit, UD-IQ1_M, 124–169 tok/s
- GeForce RTX 3090 Ti — Tight fit, UD-IQ1_M, 124–169 tok/s
- GeForce RTX 3090 — Tight fit, UD-IQ1_M, 115–157 tok/s
- NVIDIA L4 — Tight fit, UD-IQ1_M, 37–50 tok/s
- NVIDIA A10 — Tight fit, UD-IQ1_M, 74–100 tok/s
- RTX A5000 — Tight fit, UD-IQ1_M, 94–128 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-IQ4_XS).
llama.cpp
# Expert offload is approximate. llama.cpp can pin experts to CPU: # -ot "\\.ffn_.*_exps\\.=CPU" llama-server -hf unsloth/Qwen3-Coder-Next-GGUF:UD-IQ4_XS -c 8192 -ngl 99
Ollama
ollama run qwen3-coder-next
LM Studio
lms get unsloth/Qwen3-Coder-Next-GGUF # Then load quant UD-IQ4_XS