MiniMax-M2.7

Minimax family · 229.0B params

2recorded runs

495.9best tok/s

2verified

Evidence: 2 sourced runs · 1 independent source · first recorded Apr 29, 2026 · last recorded May 22, 2026. What these labels mean.

What the MiniMax-M2.7 numbers mean

The fastest verified result on our board is 495.9 output tok/s on an NVIDIA H200 SXM x4 at FP8 with a 200000-token context. That is about 2.0 ms per generated token — 500 tokens takes roughly 1.0 seconds. Peak VRAM was measured at 535.8 GB. Across the 2 records we hold for this model the range is 94.5–495.9 tok/s on 2 GPUs, because quantisation, context length and the engine build all move the number.

How much VRAM does MiniMax-M2.7 need?

MiniMax-M2.7 has 229.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.

Quantisation	Estimated VRAM	Smallest card that fits
Q4_K_M (4-bit)	148.9 GB	192 GB
Q8_0 (8-bit)	280.6 GB	multi-GPU or CPU offload
FP16 (16-bit)	527.5 GB	multi-GPU or CPU offload

Long contexts and large batches raise the real number — read the smallest card as "possible with a reduced context". Measured throughput is on the benchmark board.

Fastest GPUs for this model (best tok/s)

NVIDIA H200 SXM x4495.9
Intel Arc Pro B70 x494.5

Models with a similar footprint

Model	Params	Q4_K_M weights
Baichuan-M3-235B	235B	132.2 GB
CYBER-FROST-3.8	180B	101.2 GB
Qwen3.8-Flash-Next	180B	101.2 GB
Qwen3.8-Flash-Next-hibrid48	180B	101.2 GB
Qwen3.8-Flash-Next-REAP-256-duo	180B	101.2 GB
DeepSeek-V4-Flash	291B	163.7 GB

Ranked by distance from MiniMax-M2.7's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
NVIDIA H200 SXM x4 · 564 GB	MiniMax-M2.7	FP8	vllm 0.20.1rc1.dev30+gf05f3664c	200000	495.9 ✓	—	535.8	src
Intel Arc Pro B70 x4 · 32 GB	MiniMax-M2.7	INT4 AutoRound W4A16	vllm 0.20.1-local	4096	94.5 ✓	—	—	src