readygpu

MiniMax M3

417.1B total, 23B active · 1024k context · minimax-community · released 2026-06-02

chatcodingagenticvisionreasoning

In short

MiniMax M3 is an open model from MiniMax with 417.1B parameters (23B active per token). At mlx-4bit, MiniMax M3 needs about 227 GB of memory at 8k context and 227 GB at 8k, counting the weights, the KV cache and runtime buffers. The cheapest hardware in Germany that runs MiniMax M3 fully in memory at 8k context is the Mac Studio M3 Ultra 512GB, 18.995 € at eBay, at about 16 tok/s (estimated).

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

What to run it on in Germany

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

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 M3 Ultra 512GBUD-Q6_Kfits16 tok/s18.995 €0.8

39 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
mlx-4bitmlx224.9 GB226.6 GB229.5 GB233.5 GB241.3 GB
UD-Q4_K_Mgguf246.7 GB248.4 GB251.4 GB255.3 GB263.2 GB
UD-Q5_K_Mgguf296.8 GB298.5 GB301.5 GB305.4 GB313.3 GB
UD-Q6_Kgguf330.9 GB332.6 GB335.5 GB339.5 GB347.4 GB
Q8_0gguf422.5 GB424.1 GB427.1 GB431.0 GB438.9 GB

Run it

Copy a command for the runtime you use.

  • mlx-4bit
    • LM Studiolms get mlx-community/MiniMax-M3-4bit
    • MLXmlx_lm.server --model mlx-community/MiniMax-M3-4bit
  • UD-Q4_K_M
    • Ollamaollama run hf.co/unsloth/MiniMax-M3-GGUF:UD-Q4_K_M
    • LM Studiolms get unsloth/MiniMax-M3-GGUF@ud-q4_k_m
    • llama.cppllama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q4_K_M
  • UD-Q5_K_M
    • Ollamaollama run hf.co/unsloth/MiniMax-M3-GGUF:UD-Q5_K_M
    • LM Studiolms get unsloth/MiniMax-M3-GGUF@ud-q5_k_m
    • llama.cppllama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q5_K_M
  • UD-Q6_K
    • Ollamaollama run hf.co/unsloth/MiniMax-M3-GGUF:UD-Q6_K
    • LM Studiolms get unsloth/MiniMax-M3-GGUF@ud-q6_k
    • llama.cppllama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q6_K
  • Q8_0
    • Ollamaollama run hf.co/unsloth/MiniMax-M3-GGUF:Q8_0
    • LM Studiolms get unsloth/MiniMax-M3-GGUF@q8_0
    • llama.cppllama-server -hf unsloth/MiniMax-M3-GGUF:Q8_0

Notes

Custom "minimax-community" licence (license: other); the card doesn't state commercial-use terms, so check the LICENSE file. Supports three thinking modes: enabled, adaptive, disabled. Sparse or compressed attention: KV read per token is estimated as full attention. Excludes 9.9B parameters of multi-token-prediction layers, which local runtimes do not load. Active parameters as the card states them (23.0B); the config gives 16.9B.

Questions

How much VRAM does MiniMax M3 need?
At mlx-4bit, MiniMax M3 needs about 227 GB of memory at 8k context and 227 GB at 8k, counting the weights, the KV cache and runtime buffers.
What is the best GPU for running MiniMax M3 locally?
The cheapest hardware in Germany that runs MiniMax M3 fully in memory at 8k context is the Mac Studio M3 Ultra 512GB, 18.995 € at eBay, at about 16 tok/s (estimated).
Can I run MiniMax M3 on a 16 GB GPU?
No: it does not fit in 16 GB at 8k context, even with system RAM.
Can I run MiniMax M3 on a 24 GB GPU?
No: it does not fit in 24 GB at 8k context, even with system RAM.
How do I run MiniMax M3 locally?
With Ollama: ollama run hf.co/unsloth/MiniMax-M3-GGUF:UD-Q4_K_M (the UD-Q4_K_M build). LM Studio and llama.cpp commands are on this page.
What licence is MiniMax M3 released under?
minimax-community.