Software:Xiaomi MiMo

From HandWiki
Xiaomi MiMo
Developer(s)Xiaomi
Initial releaseApril 2025; 1 year ago (2025-04)
Stable release
MiMo-V2-Pro, MiMo-V2-Omni, MiMo-V2-TTS /
March 18, 2026; 29 days ago (2026-03-18)
PlatformCloud computing platforms
Available inEnglish and other languages
Type
LicenseProprietary
Websitemimo.xiaomi.com
mimo.mi.com

Xiaomi MiMo is a family of large language models (LLMs) developed by Xiaomi. Its first model MiMo-7B, with 7 billion parameters was launched in April, 2025.[1][2] MiMo connected billions of IoT devices in Xiaomi's "Human × Car × Home" ecosystem, intergreting its smartphones, electric vehicles and home appliances into a unified, context-aware network.[3]

Development

MiMo was developed by Xiaomi as a language model focused on reasoning. Its development team was led by Luo Fuli, who had previously worked at DeepSeek before joining Xiaomi in late 2025.[4][5] The model was trained using multi-token prediction and reinforcement learning, with a particular emphasis on mathematical reasoning and code generation tasks.[6] In March 2026, Xiaomi CEO Lei Jun announced that the company planned to invest at least US$8.7 billion in artificial intelligence over the following three years.[7]

Models

MiMo-7B

MiMo-7B is the first model of this LLM. The base model, MiMo-7B-Base, was pre-trained on approximately 25 trillion tokens using web pages, academic papers, books, and synthetic reasoning data.[6] MiMo-7B-RL underwent supervised fine-tuning and reinforcement learning on 130,000 mathematics and code problems.[6]

MiMo-7B-RL-0530 which was released in May 2025, scaled the fine-tuning dataset from 500,000 to 6 million instances and extended the RL window from 32,000 to 48,000 tokens and improved AIME 2024 scores from 68.2 to 80.1.[6]

MiMo-VL-7B was a vision-language model combining a Vision Transformer encoder with the MiMo-7B backbone. It was trained in four stages consuming 2.4 trillion tokens.[8] Its reinforcement learning variant used Mixed On-Policy Reinforcement Learning (MORL) which integrated reward signals across perception, grounding, and reasoning.[8] Xiaomi also released MiMo-Audio-7B, an audio-language model for voice conversion, style transfer, and speech editing.[9][10]

MiMo-V2-Flash

MiMo-V2-Flash was launched in December 2025. It is a open-sourced Mixture-of-Experts model with 309 billion total parameters and 15 billion active, was trained on 27 trillion tokens using FP8 mixed precision.[11][12] It used hybrid attention interleaving Sliding Window and Global Attention at a 5:1 ratio.[11]

MiMo-V2-Pro

Xiaomi publicly introduced MiMo-V2-Pro in 18 March 2026. It has over 1 trillion total parameters, 42 billion active, and a 1-million-token context window.[13] Before the official release, the model had appeared anonymously on OpenRouter under the codename "Hunter Alpha," where it drew substantial usage and topped daily charts for several days, according to Xiaomi and Reuters.[14] During its listing on OpenRouter, the model reportedly processed over one trillion tokens in total usage. Xiaomi later said Hunter Alpha was an early internal test build of MiMo-V2-Pro, and Reuters reported that the model had been mistaken by some users for a possible DeepSeek system before Xiaomi confirmed its origin.[14]

The model was released as a proprietary API product, and Luo Fuli stated that Xiaomi intended to open-source a variant at an unspecified future date. Xiaomi has partnered with several API web platforms like OpenClaw to launch the model. All these websites initially offered a free trial of this model for a week, but due to the overwhelming response, Xiaomi later extended the free trial period of the model until 2 April 2026.[15]

MiMo-V2-Omni

Alongside MiMo-V2-Pro, Xiaomi launched MiMo-V2-Omni in 18 March 2026. It handles image, video, audio, and text inputs. Before the official release, it was codenamed “Healer Alpha" in OpenRouter.[13]

MiMo-V2-TTS

On the same date as the release of MiMo-V2-Pro and MiMo-V2-Omni, a Text-to-Speech model named MiMo-V2-TTS was released also. It is a speech synthesis model. It was trained on audio data, which makes it capable of emotional transitions, mid-sentence tone shifts, singing, and synthesis of regional dialects like Sichuan, Cantonese, Henan, and Taiwanese.[13]

See also

References

  1. ↑ "Xiaomi debuts MiMo, an open-source reasoning AI model". Tech in Asia. April 30, 2025. https://www.techinasia.com/news/xiaomi-debuts-mimo-opensource-reasoning-ai-model. 
  2. ↑ "Xiaomi unveils MiMo open-source AI model for reasoning". April 30, 2025. https://www.fonearena.com/blog/452648/xiaomi-mimo-open-source-ai-model.html#:~:text=Xiaomi%20has%20introduced%20MiMo%2C%20its,edge%20devices%20with%20limited%20resources.. 
  3. ↑ "MWC 2026: Xiaomi introduces a new wave of "Human × Car × Home" driven by AI". EntrepreNerd. March 2, 2026. https://www.entnerd.com/en/mwc-2026-xiaomi-introduces-a-new-wave-of-human-car-home-driven-by-ai/?hl=en-US#:~:text=A%20fundamental%20milestone%20in%20this,developed%20by%20Xiaomi%2C%20Xiaomi%20MiMo.. 
  4. ↑ "Xiaomi strengthens AI capabilities with former DeepSeek researcher joining MiMo team". Digitimes. 16 November 2025. https://www.digitimes.com/news/a2025113PD223/xiaomi-deepseek-development-2025-language.html. 
  5. ↑ "Ex-DeepSeek Core Developer Luo Fuli Joins Xiaomi to Lead MiMo Team on Spatial Intelligence". Pandaily. 12 November 2025. https://pandaily.com/ex-deep-seek-core-developer-luo-fuli-joins-xiaomi-to-lead-mi-mo-team-on-spatial-intelligence. 
  6. ↑ 6.0 6.1 6.2 6.3 Xiaomi (12 May 2025). "MiMo: Unlocking the Reasoning Potential of Language Model — From Pretraining to Posttraining". arXiv:2505.07608 [cs.CL].
  7. ↑ "Xiaomi to invest at least $8.7 billion in AI over next three years, CEO says". Reuters. 19 March 2026. https://www.reuters.com/world/asia-pacific/xiaomi-invest-least-87-billion-ai-over-next-three-years-ceo-says-2026-03-19/. 
  8. ↑ 8.0 8.1 Xiaomi (4 June 2025). "MiMo-VL Technical Report". arXiv:2506.03569 [cs.CV].
  9. ↑ Xiaomi LLM-Core Team (29 December 2025). "MiMo-Audio: Audio Language Models are Few-Shot Learners". arXiv:2512.23808 [cs.SD].
  10. ↑ "Mimo-audio Enables Few-Shot Learning for Audio Tasks, Scaling to 100 Million Hours". Quantam Zeitgeist. 8 January 2026. https://quantumzeitgeist.com/learning-mimo-audio-enables-few-shot/. 
  11. ↑ 11.0 11.1 Xiaomi LLM-Core Team (6 January 2026). "MiMo-V2-Flash Technical Report". arXiv:2601.02780 [cs.CL].
  12. ↑ Ben Jiang (17 December 2025). "Xiaomi launches open-source model to compete against DeepSeek, Moonshot, OpenAI". South China Morning Post. https://www.scmp.com/tech/tech-trends/article/33672/xiaomi-launches-open-source-model-compete-against-deepseek-moonshot-openai-systems. 
  13. ↑ 13.0 13.1 13.2 "Xiaomi Launches Powerful AI Model MiMo-V2 Pro With 1 Trillion Parametres, 1 Million Token Context Window". NDTV Profit. 19 March 2026. https://www.ndtvprofit.com/technology/xiaomi-launches-powerful-ai-model-mimo-v2-pro-with-1-trillion-parametres-1-million-token-context-window-11236705. 
  14. ↑ 14.0 14.1 "Mystery AI model revealed to be Xiaomi's following suspicions it was DeepSeek's". Reuters. 18 March 2026. https://www.reuters.com/business/media-telecom/mystery-ai-model-has-developers-buzzing-is-this-deepseets-latest-blockbuster-2026-03-18/. 
  15. ↑ "Xiaomi Extends MiMo-V2 Agent Framework Free Trial". 26 March 2026. https://xiaomitime.com/xiaomi-extends-mimo-v2-agent-framework-free-trial-94302/. 



Retrieved from "https://handwiki.org/wiki/index.php?title=Software:Xiaomi_MiMo&oldid=4298637"