Xiaomi has released its MiMo family of large language models, featuring a trillion parameters in its latest MiMo-V2.5-Pro version. These models are notable for their open-weight status under the MIT license and their native omnimodality, capable of processing text, images, video, and audio. The MiMo-V2.5-Pro utilizes a sparse mixture-of-experts architecture with 42 billion active parameters per token out of a total of 1.02 trillion, and supports a context window of up to 1 million tokens. AI
IMPACT This release signals that open-weight models are now capable of reaching trillion-parameter scale, potentially democratizing access to advanced AI capabilities.
RANK_REASON The cluster details the release of a new, large-scale language model family by a major tech company, including specific parameter counts and architectural details. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- GraphWalks
- MiMo
- MiMo-7B
- MiMo-Audio
- MiMo-Embodied
- MiMo-V2.5
- MiMo-V2.5-ASR
- MiMo-V2.5-Pro
- MiMo-VL
- MXFP4
- Xiaomi
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →