readygpu

Mistral Small 4 119B

119.4B total, 7.1B active · 1024k context · apache-2.0 · released 2026-01-23

chatcodingagenticreasoningvision

What to run it on in the US

Fully in memory at 128k 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-Q5_K_Mfits26 tok/s––
Mac mini M4 Pro 64GBUD-Q2_K_XLfits20 tok/s––
Framework Desktop Ryzen AI Max+ 395 128GBUD-Q5_K_Mfits10 tok/s––
Mac Studio M3 Ultra 512GBQ8_0fits30 tok/s––
NVIDIA DGX Spark 128GBUD-Q6_Kfits11 tok/s––
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBUD-Q5_K_Mfits10 tok/s––
Mac Studio M3 Ultra 256GBQ8_0fits30 tok/s––
Mac Studio M4 Max 64GBUD-Q2_K_XLfits40 tok/s––
Mac Studio M3 Ultra 96GBUD-Q3_K_Mfits51 tok/s––
NVIDIA RTX PRO 6000 BlackwellUD-Q5_K_Mfits80 tok/s––
ASUS Ascent GX10 128GBUD-Q6_Kfits11 tok/s––
Framework Desktop Ryzen AI Max+ 395 64GBUD-Q2_K_XLfits16 tok/s––
NVIDIA GeForce RTX 3090UD-Q2_K_XLoffload12 tok/s––
NVIDIA GeForce RTX 5090UD-Q2_K_XLoffload21 tok/s––
NVIDIA GeForce RTX 4090UD-Q2_K_XLoffload12 tok/s––
AMD Radeon RX 7900 XTXUD-Q2_K_XLoffload12 tok/s––
AMD Radeon AI PRO R9700UD-Q2_K_XLoffload16 tok/s––
AMD Radeon RX 7900 XTUD-Q2_K_XLoffload10 tok/s––
NVIDIA GeForce RTX 3090 TiUD-Q2_K_XLoffload12 tok/s––
Intel Arc Pro B60UD-Q2_K_XLoffload9 tok/s––
NVIDIA Tesla P40UD-Q2_K_XLoffload10 tok/s––

19 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_Mgguf68.7 GB69.6 GB70.3 GB71.2 GB73.0 GB
UD-Q2_K_XLgguf37.4 GB38.4 GB39.0 GB39.9 GB41.7 GB
UD-Q3_K_Mgguf50.6 GB51.6 GB52.2 GB53.1 GB54.9 GB
mlx-4bitmlx63.1 GB64.0 GB64.7 GB65.6 GB67.4 GB
UD-Q5_K_Mgguf83.0 GB84.0 GB84.6 GB85.5 GB87.3 GB
UD-Q6_Kgguf92.6 GB93.5 GB94.2 GB95.1 GB96.9 GB
Q8_0gguf117.8 GB118.7 GB119.4 GB120.3 GB122.0 GB
mlx-8bitmlx118.5 GB119.4 GB120.1 GB121.0 GB122.8 GB

Run it

Copy a command for the runtime you use.

  • UD-Q4_K_M
    • Ollamaollama run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q4_K_M
    • LM Studiolms get unsloth/Mistral-Small-4-119B-2603-GGUF@ud-q4_k_m
    • llama.cppllama-server -hf unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q4_K_M
  • UD-Q2_K_XL
    • Ollamaollama run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q2_K_XL
    • LM Studiolms get unsloth/Mistral-Small-4-119B-2603-GGUF@ud-q2_k_xl
    • llama.cppllama-server -hf unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q2_K_XL
  • UD-Q3_K_M
    • Ollamaollama run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q3_K_M
    • LM Studiolms get unsloth/Mistral-Small-4-119B-2603-GGUF@ud-q3_k_m
    • llama.cppllama-server -hf unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q3_K_M
  • mlx-4bit
    • LM Studiolms get mlx-community/Mistral-Small-4-119B-2603-4bit
    • MLXmlx_lm.server --model mlx-community/Mistral-Small-4-119B-2603-4bit
  • UD-Q5_K_M
    • Ollamaollama run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q5_K_M
    • LM Studiolms get unsloth/Mistral-Small-4-119B-2603-GGUF@ud-q5_k_m
    • llama.cppllama-server -hf unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q5_K_M
  • UD-Q6_K
    • Ollamaollama run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q6_K
    • LM Studiolms get unsloth/Mistral-Small-4-119B-2603-GGUF@ud-q6_k
    • llama.cppllama-server -hf unsloth/Mistral-Small-4-119B-2603-GGUF:UD-Q6_K
  • Q8_0
    • Ollamaollama run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:Q8_0
    • LM Studiolms get unsloth/Mistral-Small-4-119B-2603-GGUF@q8_0
    • llama.cppllama-server -hf unsloth/Mistral-Small-4-119B-2603-GGUF:Q8_0
  • mlx-8bit
    • LM Studiolms get mlx-community/Mistral-Small-4-119B-2603-8bit
    • MLXmlx_lm.server --model mlx-community/Mistral-Small-4-119B-2603-8bit