Xiaomi Unveils Open-Source MiMo-V2-Flash AI Model to Challenge DeepSeek and OpenAI
Xiaomi has launched a new open-source artificial intelligence model, stepping up competition with leading systems from DeepSeek, Moonshot AI, Anthropic, and OpenAI as China’s AI race accelerates.
The Chinese smartphone and electric vehicle maker said its latest model, MiMo-V2-Flash, is designed to deliver strong performance in reasoning, coding, and agent-based tasks, while also functioning as a capable general-purpose assistant. The model is now available globally through MiMo Studio, Hugging Face, and Xiaomi’s API Platform.
Built for Speed, Reasoning, and Low Cost
According to Xiaomi, MiMo-V2-Flash was engineered for efficiency. The company claims the model can process 150 tokens per second, while keeping costs extremely low at $0.10 per million input tokens and $0.30 per million output tokens, positioning it among the most cost-effective high-performance AI models currently available.
Luo Fuli, a prominent Chinese AI researcher who recently joined Xiaomi’s MiMo team after working at DeepSeek, described the release as “step two on our AGI roadmap” in a post on X. AGI, or artificial general intelligence, refers to AI systems capable of matching or exceeding human-level cognitive abilities.
Competitive Benchmark Results
Xiaomi said MiMo-V2-Flash delivers performance comparable to Moonshot AI’s Kimi K2 Thinking and DeepSeek V3.2 Thinking across most reasoning benchmarks, while maintaining strong general writing capabilities.
In long-context evaluations, Xiaomi claims its model outperformed Kimi K2 Thinking. On agent-based tasks, MiMo-V2-Flash reportedly scored 73.4% on SWE-Bench Verified, surpassing all open-source competitors and approaching the performance of OpenAI’s GPT-5-High.
The model was also said to match the coding performance of Claude 4.5 Sonnet, widely regarded as one of the strongest coding-focused AI models, but at a fraction of the cost.
Architecture and Technical Design
MiMo-V2-Flash is built using a Mixture-of-Experts architecture, which divides the model into specialized sub-networks that activate selectively depending on the task. The system contains 309 billion total parameters, enabling it to handle complex reasoning and long-context workloads efficiently.
AI engineer Rohan Paul noted that the design significantly reduces the cost and time required to process long prompts, as the model avoids comparing every token with all previous inputs.
Strategic Importance for Xiaomi
Analysts say the release comes at a critical moment as Xiaomi pushes deeper into AI-powered applications across smartphones, tablets, electric vehicles, and robotics.
Xiaomi president Lu Weibing recently said the company’s progress in large AI models has exceeded expectations and confirmed a strategic pivot toward embodied AI, applying intelligence directly to physical devices such as EVs and robots.
With MiMo-V2-Flash gaining attention globally, Xiaomi is positioning itself as a serious contender in the rapidly evolving AI landscape—well beyond its roots in consumer electronics
Search us on Google it will help us
0 Comments