LocalOps
Back to Calculator

MiniMax M2.7

MiniMax M2.7 is a 230B sparse MoE model built for complex agentic workflows. 10B parameters active per token. Open weights under non-commercial license. Released March 2026.

Model Specifications

ArchitectureTEXT
Parameters230B
Familyminimax
VRAM (Q4)115.0GB
Mixture of ExpertsActive inference parameters: 10B.
#reasoning#agents#codingSource

Estimated Quantization Sizes

FormatPrecisionEst. VRAMRecommendation
FP16 / BF1616-bit460.0 GBUncompressed Base
Q8_0High8-bit230.0 GBNear Lossless
Q6_K6-bit172.5 GBExcellent Balance
Q4_K_MPopular4-bit115.0 GBStandard Use

Share this Model

Send this model's specs directly to your community.

Post