readygpu

Qwen 3.8 Flash Next

180B total, 61.6B active · 256k context · qwen-community-1.0 · released 2026-08-24

chatcodingagenticreasoningvision

What to run it on in the UK

Fully in memory at 8k context, priced at retailers today.

Something we price holds it fully in memory, but none reaches 15 tok/s, our bar for recommending it. The table below has the speeds.

All hardware that runs it

Everything we price, best value first. Two cards split the model by layers.

HardwareQuantFitsDecodePricetok/s per 1,000
Mac Studio M4 Max 128GBUD-Q3_K_XLfits7 tok/s––
Framework Desktop Ryzen AI Max+ 395 128GBUD-Q3_K_XLfits3 tok/s––
Mac Studio M3 Ultra 512GBQ8_0fits5 tok/s––
NVIDIA DGX Spark 128GBUD-Q4_K_XLfits3 tok/s––
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBUD-Q3_K_XLfits3 tok/s––
Mac Studio M3 Ultra 256GBQ8_0fits5 tok/s––
NVIDIA RTX PRO 6000 BlackwellUD-Q3_K_XLfits21 tok/s––
ASUS Ascent GX10 128GBUD-Q4_K_XLfits3 tok/s––

32 other cards and machines we price cannot hold it at this context.

Memory needed

Weights, KV cache and runtime buffers at each context length (KV cache at fp16).

QuantFile@8k@32k@64k@128k
UD-Q4_K_XLgguf103.7 GB104.7 GB105.4 GB106.4 GB108.2 GB
UD-Q2_K_XLgguf73.5 GB74.5 GB75.2 GB76.1 GB78.0 GB
UD-Q3_K_XLgguf83.8 GB84.8 GB85.5 GB86.5 GB88.4 GB
mlx-4bitmlx103.9 GB104.9 GB105.6 GB106.5 GB108.4 GB
UD-Q5_K_XLgguf147.4 GB148.5 GB149.2 GB150.1 GB152.0 GB
UD-Q6_K_XLgguf157.5 GB158.6 GB159.3 GB160.2 GB162.1 GB
Q8_0gguf175.3 GB176.3 GB177.0 GB178.0 GB179.9 GB

Run it

Copy a command for the runtime you use.

  • UD-Q4_K_XL
    • Ollamaollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL
    • LM Studiolms get unsloth/Qwen3.8-Flash-Next-GGUF@ud-q4_k_xl
    • llama.cppllama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL
  • UD-Q2_K_XL
    • Ollamaollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q2_K_XL
    • LM Studiolms get unsloth/Qwen3.8-Flash-Next-GGUF@ud-q2_k_xl
    • llama.cppllama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q2_K_XL
  • UD-Q3_K_XL
    • Ollamaollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q3_K_XL
    • LM Studiolms get unsloth/Qwen3.8-Flash-Next-GGUF@ud-q3_k_xl
    • llama.cppllama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q3_K_XL
  • mlx-4bit
    • LM Studiolms get mlx-community/Qwen3.8-Flash-Next-4bit
    • MLXmlx_lm.server --model mlx-community/Qwen3.8-Flash-Next-4bit
  • UD-Q5_K_XL
    • Ollamaollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q5_K_XL
    • LM Studiolms get unsloth/Qwen3.8-Flash-Next-GGUF@ud-q5_k_xl
    • llama.cppllama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q5_K_XL
  • UD-Q6_K_XL
    • Ollamaollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q6_K_XL
    • LM Studiolms get unsloth/Qwen3.8-Flash-Next-GGUF@ud-q6_k_xl
    • llama.cppllama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q6_K_XL
  • Q8_0
    • Ollamaollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:Q8_0
    • LM Studiolms get unsloth/Qwen3.8-Flash-Next-GGUF@q8_0
    • llama.cppllama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:Q8_0

Notes

Estimates are approximate. Mixture-of-experts parameters do not add up; active size is approximate.