Skip to main content
Source: Xiaomi MiMo platform. All three models support tool/function calling.

​
MiMo V2

mimo-v2-pro

Reasoning · Speedxiaomi/mimo-v2-proXiaomi’s flagship reasoning model with ~1 trillion parameters (42B active, MoE). Approaches Claude Opus 4.6 on agent and coding benchmarks at a fraction of the cost. Supports extended thinking via <think> tags.
  • $1 / $3
  • 1M context
  • 32K max output
  • Text input
  • Thinking (<think> tags)

mimo-v2-omni

Reasoning · Speedxiaomi/mimo-v2-omniMultimodal reasoning model capable of natively processing text, images, video, and over 10 hours of continuous audio. Designed for agentic workflows requiring perception across modalities.
  • $0.40 / $2
  • 256K context
  • 32K max output
  • Text, Image, Video, Audio input

mimo-v2-flash

Reasoning · Speedxiaomi/mimo-v2-flashLightweight open-source (MIT) MoE model with 309B total parameters and 15B active. Optimized for high-speed reasoning and agentic workflows via a hybrid attention architecture.
  • $0.09 / $0.29
  • 262K context
  • Text input
  • Open-source (MIT)