ModelMIMO V2.6
MIMO V2.6 Flash
An open-weight large language model released by Xiaomi, built on a Mixture of Experts (MoE) architecture with approximately 309 billion total parameters and roughly 15 billion parameters activated per token. It is designed for coding and agent tasks.
Timeline (last 90 days)
Oct 5
MIMO under MXFP4 starts prompt processing at approximately 477 tokens/s, dropping to 354 at 32K, with token generation declining from 18 to 16.
Expired50%