Skip to content

How much VRAM does MiniMax M3 need?

MiniMax M3 needs 241 GB for its weights at 4-bit, 243 GB with an 8K context and 246 GB with 32K. At 8-bit, 425 GB.

Parameters
427B
Active per token
23B
Longest context
1M
Released
Jun 2026
License
MiniMax Community
Strongest at
Docs

Memory at each precision

Memory MiniMax M3 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit795 GB797 GB800 GB811 GB
8-bit423 GB425 GB427 GB439 GB
6-bit326 GB328 GB331 GB342 GB
5-bit283 GB285 GB288 GB299 GB
4-bit241 GB243 GB246 GB257 GB
3-bit194 GB196 GB199 GB210 GB
2-bit149 GB151 GB154 GB165 GB
1.6-bit82.0 GB84.0 GB86.8 GB98.0 GB

One conversation, 16-bit context cache, runtime overhead included.

What runs it

The best way MiniMax M3 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst reply
RTX 3060 12 GBToo big
RTX 4060 Ti 16 GBToo big
RTX 3090Too big
RTX 4090Too big
RTX 5090Too big
2× RTX 3090Too big
RTX PRO 6000 Blackwell1.6-bit64K135 tok/s1.2 s
M4 Max 128 GB1.6-bit64K46 tok/s24 s
M5 Max 128 GB1.6-bit64K49 tok/s7.0 s
Ryzen AI Max+ 395 128 GB1.6-bit64K28 tok/s14 s
DGX Spark 128 GB1.6-bit256K35 tok/s4.9 s
M3 Ultra 256 GB2-bit256K41 tok/s14 s
M5 Ultra 512 GB6-bit320K31 tok/s4.3 s
H100 SXMToo big
DGX Station GB3003-bit352K171 tok/s0.4 s
8× H200 NVL8-bit1M92 tok/s0.2 s

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF
vLLM
vllm serve MiniMaxAI/MiniMax-M3
MLX (Mac)
mlx_lm.chat --model mlx-community/MiniMax-M3-4bit

Questions

How much VRAM does MiniMax M3 need?

MiniMax M3 needs 241 GB for its weights at 4-bit, 243 GB with an 8K context and 246 GB with 32K. At 8-bit, 425 GB.

Can MiniMax M3 run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the RTX PRO 6000 Blackwell.

More models