Xiaomi

A 38 billion parameter multimodal autoregressive model for unified embodied synthesis. It performs text to image generation, image editing, multi view robot scene generation, embodied scene transfer, and embodied video…

• 🤖 Robotics
• 🖼️ Image generation
• 🖌️ Image editing
• 🔀 Synthetic data
• 🎥 Videos New

Xiaomi MiMo’s 1T-parameter high-speed reasoning model variant built with TileRT to reach 1000+ tokens per second generation on commodity GPU infrastructure.

• 💻 Coding
• 🔍 Code reviews
• ✉️ Codebase Q&A

Xiaomi MiMo’s state-of-the-art end-to-end ASR model for robust transcription across Mandarin, English, Chinese dialects, code-switching, songs, noisy audio, and multi-speaker conversations. It is an 8B model aimed at di…

• 🗒 Transcription
• 🎙️ Voice recognition
• 🎤 Voice transcription

Xiaomi’s latest agentic language model, built for stronger long-horizon coherence, complex software engineering, and autonomous tool-using workflows. It is positioned as Xiaomi’s most capable model so far, with better i…

• 💻 Coding
• 🔍 Code reviews
• ✉️ Codebase Q&A
• 🤖 Agents
• 🎩 Sophisticated reasoning

A multilingual zero-shot text-to-speech model built for voice cloning, voice design, and general speech synthesis at massive language scale. It supports more than 600 languages, uses a diffusion language model-style arc…

• 🔊 Text to speech
• 🎤 Voice changing
• 🗣️ Voice cloning
• 🌐 Multilingual communication

MiMo MiMo-V2-TTS is Xiaomi’s large-scale speech synthesis model built for expressive agent voice, aiming for natural, emotionally aware speech.

• 🔊 Text to speech
• 🎤 Voice changing
• 🎙️ Voiceovers
• 🎤 Singing

MiMo MiMo-V2-Omni is an omni foundation model that unifies multimodal understanding with agentic capability, built to see, hear, and act.

• 📚 Large language models
• 🎥 Video summaries
• 📚 Audio summaries
• 🔍 Image interpretation

MiMo MiMo-V2-Pro is Xiaomi’s flagship foundation model built for real-world agent workloads, designed to act as the “brain” of agent systems that orchestrate complex workflows and tool use.

• 📚 Large language models
• 🤖 Agents
• 🔄 Workflow optimization
• 💻 Conversational coding

A 4.7B-parameter open Vision-Language-Action model that uses a Mixture-of-Transformers design, combining a Qwen3-based vision-language brain with a diffusion transformer controller for smooth, real-time robot manipulati…

• 🔒 Private conversations
• 📝 Lecture summaries
• 🖼️ Blog images

A 4.7B parameter vision language action model fine tuned on the LIBERO benchmark. It takes camera images, a language instruction, and robot proprioceptive state as input and outputs action chunks to control a bimanual r…

• 🤖 Robotics
• 🤖 Robotics guidance

MiMo I cannot find public technical documentation for a distinct “MiMo v2 Flash” model beyond Xiaomi’s MiMo-7B and MiMo-VL releases, so I cannot reliably describe that specific variant without guessing.