Home
Xiaomi Unveils MiMo-V2.5, Its Most Powerful Model Series, as Official Public Testing Begins
On April 23 at 04:23, Xiaomi advanced its AI strategy with the official launch of the MiMo-V2.5 series, simultaneously initiating public beta testing. This release represents a fundamental technical leap rather than a minor update. The lineup features four distinct models: MiMo-V2.5, V2.5-Pro, V2.5-TTS Series, and V2.5-ASR. Notably, Xiaomi has globally open-sourced the core MiMo-V2.5-Pro and MiMo-V2.5 models, signaling a strong commitment to fostering an open AI ecosystem and attracting developer engagement.


Flagship Performance: Mastering Million-Token Contexts and Complex Workflows
Positioned as the premium tier, MiMo-V2.5-Pro is engineered for intricate Agent tasks requiring long-context processing. It supports an impressive 1 million Token context window, delivering exceptional results in software engineering and complex logical operations. In a benchmark involving the SysY compiler project from Peking University’s “Principles of Compilation” course, the model completed a task typically taking undergraduates several weeks in just 4.3 hours, utilizing 672 tool calls and achieving a perfect score. This capacity for sustained logical consistency allows it to compete directly with leading closed-source models such as Claude Opus 4.6 and GPT-5.4.

Full-Modal Versatility: High Efficiency and Broad Utility
The standard MiMo-V2.5 model emphasizes native full-modal capabilities, seamlessly integrating image, audio, and video processing. It has demonstrated superior reasoning speed and outperformed its Pro counterpart in authoritative Agent evaluation benchmarks. For developers, a key advantage is the approximately 50% reduction in API call costs achieved through technical optimization. In areas like cross-modal reasoning, video understanding, and professional chart analysis, MiMo-V2.5 matches or exceeds industry-leading models in specific metrics.

Optimized Pricing: Maximizing Output per Token
Beyond raw performance, Xiaomi has significantly enhanced underlying efficiency. Data indicates that the MiMo-V2.5 series consumes fewer Tokens than competitors across identical testing benchmarks. To support this launch, Xiaomi has revamped its “Token Plan” subscription structure, removing billing disparities based on context length and eliminating multi-tiered pricing. Additionally, the introduction of nighttime discounted rates and auto-renewal options further lowers the barrier to entry for enterprises and developers adopting large language models.
Related article
TikTok launches verifiable rights protection mechanism to counter AI-synthesized impersonation
On September 17, Douyin launched a dedicated reporting channel for "voice impersonation" and upgraded its rights protection mechanisms, addressing new demands driven by the widespread adoption of AI generation tools.With lower barriers to AI generati
OpenAI Exec Lehan: AI Can Now Plan Cyber Attacks; Public Firms Must Prepare Defenses
Chris Liuhan, OpenAI’s Chief Global Affairs Officer, cautioned The Guardian that state-of-the-art AI models can now orchestrate sophisticated cyberattacks, urging businesses and the public to brace for relentless threats. He emphasized that AI has tr
Meta Wants Your Next Gadget to Be Muse-Infused
Meta’s Muse, an AI assistant that handles travel bookings, form submissions, and shopping, has gained widespread adoption, yet a new initiative may especially intrigue developers and hardware enthusiasts.On Friday, Meta unveiled Muse Gadgets, an open
Related Special Topic Recommendations
Comments (0)
0/500
On April 23 at 04:23, Xiaomi advanced its AI strategy with the official launch of the MiMo-V2.5 series, simultaneously initiating public beta testing. This release represents a fundamental technical leap rather than a minor update. The lineup features four distinct models: MiMo-V2.5, V2.5-Pro, V2.5-TTS Series, and V2.5-ASR. Notably, Xiaomi has globally open-sourced the core MiMo-V2.5-Pro and MiMo-V2.5 models, signaling a strong commitment to fostering an open AI ecosystem and attracting developer engagement.


Flagship Performance: Mastering Million-Token Contexts and Complex Workflows
Positioned as the premium tier, MiMo-V2.5-Pro is engineered for intricate Agent tasks requiring long-context processing. It supports an impressive 1 million Token context window, delivering exceptional results in software engineering and complex logical operations. In a benchmark involving the SysY compiler project from Peking University’s “Principles of Compilation” course, the model completed a task typically taking undergraduates several weeks in just 4.3 hours, utilizing 672 tool calls and achieving a perfect score. This capacity for sustained logical consistency allows it to compete directly with leading closed-source models such as Claude Opus 4.6 and GPT-5.4.

Full-Modal Versatility: High Efficiency and Broad Utility
The standard MiMo-V2.5 model emphasizes native full-modal capabilities, seamlessly integrating image, audio, and video processing. It has demonstrated superior reasoning speed and outperformed its Pro counterpart in authoritative Agent evaluation benchmarks. For developers, a key advantage is the approximately 50% reduction in API call costs achieved through technical optimization. In areas like cross-modal reasoning, video understanding, and professional chart analysis, MiMo-V2.5 matches or exceeds industry-leading models in specific metrics.

Optimized Pricing: Maximizing Output per Token
Beyond raw performance, Xiaomi has significantly enhanced underlying efficiency. Data indicates that the MiMo-V2.5 series consumes fewer Tokens than competitors across identical testing benchmarks. To support this launch, Xiaomi has revamped its “Token Plan” subscription structure, removing billing disparities based on context length and eliminating multi-tiered pricing. Additionally, the introduction of nighttime discounted rates and auto-renewal options further lowers the barrier to entry for enterprises and developers adopting large language models.
TikTok launches verifiable rights protection mechanism to counter AI-synthesized impersonation
On September 17, Douyin launched a dedicated reporting channel for "voice impersonation" and upgraded its rights protection mechanisms, addressing new demands driven by the widespread adoption of AI generation tools.With lower barriers to AI generati
OpenAI Exec Lehan: AI Can Now Plan Cyber Attacks; Public Firms Must Prepare Defenses
Chris Liuhan, OpenAI’s Chief Global Affairs Officer, cautioned The Guardian that state-of-the-art AI models can now orchestrate sophisticated cyberattacks, urging businesses and the public to brace for relentless threats. He emphasized that AI has tr
Meta Wants Your Next Gadget to Be Muse-Infused
Meta’s Muse, an AI assistant that handles travel bookings, form submissions, and shopping, has gained widespread adoption, yet a new initiative may especially intrigue developers and hardware enthusiasts.On Friday, Meta unveiled Muse Gadgets, an open











