FAMILY / XIAOMI MIMO

MiMo

Xiaomi’s family spanning a 7B reasoning model through trillion-parameter, multimodal, 1M-context agents.

First model: 2025
Models: 11
Latest dated version: MiMo-V2.6-Pro-UltraSpeed

01 / RELEASE SEQUENCE

Family release sequence

Compare versions →

02 / GENERATIONS

What changed

MiMo-7B-RL

Xiaomi’s first open reasoning model, exploring the full pretraining-to-RL path at 7B scale.

2025-04-30

MiMo-V2-Flash

Used hybrid sliding-window attention and MTP for faster long-context coding and agents.

2025-12-16

MiMo-V2-Omni

Unified text, vision, and speech perception with tools and GUI control in one agent foundation.

2026-03-18

MiMo-V2-Pro

A trillion-parameter, 42B-active agent flagship with a 1M context window.

2026-03-18

MiMo-V2.5

MiMo’s current general model combining native full-modal understanding, 1M context, and agent execution.

2026-04-23

MiMo-V2.5-Pro

The high-intensity MiMo-V2.5 agent tier for long-range reasoning, coding, and complex-task efficiency.

2026-04-23

MiMo-X Preview

This is MiMo’s preview step from open models and APIs toward long-chain desktop delivery; Xiaomi has not announced either Preview tier as public GA or open weights.

2026-09-08

MiMo-V2.6 turns the previously public reinforcement-learning runs into a formal API, desktop product, and full open-weight release, while also sharing RL environments, the training framework, and Distill-Qwen-9B resources.

2026-09-21

MiMo-V2.6-Flash

Flash brings V2.6 native multimodality and RL capability to high-throughput APIs and everyday multimodal workflows.

2026-09-21

MiMo-V2.6-Pro

Pro exposes MiMo-V2.6’s large-scale RL post-training as a standalone flagship API and open-weight model.

2026-09-21

MiMo-V2.6-Pro-UltraSpeed

UltraSpeed records Pro’s capability with a separate serving-speed tier, avoiding a claim of a new base-model weight release.

2026-09-21