readygpu

Mistral Medium 3.5 128B

127.7B parameters · 256k context · Modified MIT License · released 2026-03-31

chatcodingagenticvisionreasoning

In short

Mistral Medium 3.5 128B is an open model from Mistral AI with 127.7B parameters. At Q4_K_M, Mistral Medium 3.5 128B needs about 73 GB of memory at 8k context and 73 GB at 8k, counting the weights, the KV cache and runtime buffers. Nothing we price in the US runs Mistral Medium 3.5 128B fully in memory at 8k context at a usable speed; see the table for what comes closest.

Updated . Speeds are estimates; how they are worked out.

What to run it on in the US

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
Framework Desktop Ryzen AI Max+ 395 128GBQ5_K_Mfits2 tok/s$3,4490.4
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBQ5_K_Mfits2 tok/s$3,6500.4
Mac Studio M4 Max 128GBQ5_K_Mfits4 tok/s$4,9990.8
ASUS Ascent GX10 128GBQ6_Kfits2 tok/s$6,1000.3
NVIDIA DGX Spark 128GBQ6_Kfits2 tok/s$7,3990.2
NVIDIA RTX PRO 6000 BlackwellQ5_K_Mfits12 tok/s$7,9971.5
Mac Studio M3 Ultra 256GBQ8_0fits4 tok/s$8,4990.5
Mac Studio M3 Ultra 512GBQ8_0fits4 tok/s$23,0000.2

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
Q4_K_Mgguf69.8 GB73.3 GB81.6 GB92.8 GB115.2 GB
Q5_K_Mgguf82.3 GB85.7 GB94.1 GB105.3 GB127.7 GB
Q6_Kgguf95.5 GB99.0 GB107.4 GB118.6 GB141.0 GB
Q8_0gguf123.7 GB127.2 GB135.6 GB146.8 GB169.2 GB

Run it

Copy a command for the runtime you use.

  • Q4_K_M
    • Ollamaollama run hf.co/unsloth/Mistral-Medium-3.5-128B-GGUF:Q4_K_M
    • LM Studiolms get unsloth/Mistral-Medium-3.5-128B-GGUF@q4_k_m
    • llama.cppllama-server -hf unsloth/Mistral-Medium-3.5-128B-GGUF:Q4_K_M
  • Q5_K_M
    • Ollamaollama run hf.co/unsloth/Mistral-Medium-3.5-128B-GGUF:Q5_K_M
    • LM Studiolms get unsloth/Mistral-Medium-3.5-128B-GGUF@q5_k_m
    • llama.cppllama-server -hf unsloth/Mistral-Medium-3.5-128B-GGUF:Q5_K_M
  • Q6_K
    • Ollamaollama run hf.co/unsloth/Mistral-Medium-3.5-128B-GGUF:Q6_K
    • LM Studiolms get unsloth/Mistral-Medium-3.5-128B-GGUF@q6_k
    • llama.cppllama-server -hf unsloth/Mistral-Medium-3.5-128B-GGUF:Q6_K
  • Q8_0
    • Ollamaollama run hf.co/unsloth/Mistral-Medium-3.5-128B-GGUF:Q8_0
    • LM Studiolms get unsloth/Mistral-Medium-3.5-128B-GGUF@q8_0
    • llama.cppllama-server -hf unsloth/Mistral-Medium-3.5-128B-GGUF:Q8_0

Notes

The original Transformers config had a bug that degraded long-context performance, and GGUFs made before the fix commit are also affected. The vLLM recipe uses tensor-parallel-size 8 and needs vLLM nightly, mistral_common >= 1.11.1 and transformers >= 5.4.0.

Questions

How much VRAM does Mistral Medium 3.5 128B need?
At Q4_K_M, Mistral Medium 3.5 128B needs about 73 GB of memory at 8k context and 73 GB at 8k, counting the weights, the KV cache and runtime buffers.
Can I run Mistral Medium 3.5 128B on a 16 GB GPU?
No: it does not fit in 16 GB at 8k context, even with system RAM.
Can I run Mistral Medium 3.5 128B on a 24 GB GPU?
No: it does not fit in 24 GB at 8k context, even with system RAM.
How do I run Mistral Medium 3.5 128B locally?
With Ollama: ollama run hf.co/unsloth/Mistral-Medium-3.5-128B-GGUF:Q4_K_M (the Q4_K_M build). LM Studio and llama.cpp commands are on this page.
What licence is Mistral Medium 3.5 128B released under?
Modified MIT License.