Chinese tech giant Xiaomi has made a significant leap into the competitive world of artificial intelligence with the launch of its latest open-weight model, MiMo-V2-Flash. Announced on December 20, 2025, this model is engineered to handle sophisticated tasks like complex reasoning, coding, and powering AI agents, positioning Xiaomi as a direct competitor to established players like DeepSeek, Anthropic, and OpenAI.
Technical Prowess and Competitive Pricing
The newly unveiled MiMo-V2-Flash is a behemoth with a total of 309 billion parameters, a key indicator of its processing power and scale. Xiaomi has emphasized its efficiency, stating the model can deliver inference speeds of up to 150 tokens per second. More importantly, it operates at a remarkably low cost, priced at $0.1 per million input tokens and $0.3 per million output tokens.
Architecturally, the model employs a Mixture-of-Experts (MoE) framework. This design splits large neural networks into smaller, specialized "experts," allowing the system to balance top-tier performance with computational efficiency. A key innovation is its approach to handling long prompts, which reduces costs by limiting the amount of past context the model needs to re-evaluate during processing.
Benchmark Performance and Strategic Vision
In terms of raw performance, Xiaomi claims MiMo-V2-Flash holds its own against the best in the industry. The company reports that the model's scores are on par with Moonshot AI's Kimi K2 Thinking and DeepSeek V3.2 in most reasoning tests. It even surpassed Kimi K2 in long-context evaluations.
For coding tasks, the model reportedly matched the performance of Anthropic's Claude 4.5 Sonnet, a notable achievement given Xiaomi's claim that MiMo-V2-Flash was built at a fraction of the cost. On the SWE-Bench Verified benchmark, it outperformed all open AI model rivals with a score of 73.4%.
The model is now publicly accessible for developers via Xiaomi's MiMo Studio portal, the Hugging Face platform, and through its dedicated API. This release is the second step in what Xiaomi describes as its AGI (Artificial General Intelligence) roadmap. Luo Fuli, a former DeepSeek researcher now on Xiaomi's MiMo team, highlighted the engineering choices behind the model in a social media post on December 17, 2025.
Integrating AI Across Xiaomi's Ecosystem
The launch is not an isolated event but a core part of Xiaomi's broader strategy. The company is actively working to integrate AI agent-driven features across its entire product portfolio, including its popular smartphones, tablets, and its burgeoning electric vehicle (EV) line. This move signals a strategic expansion beyond hardware manufacturing into the foundational layer of AI technology.
Lu Weibing, President of Xiaomi, recently noted that the company's progress in AI large models has "far exceeded our expectations." He emphasized Xiaomi's belief that the deep integration of AI with the physical world represents the next major frontier in technology. The introduction of MiMo-V2-Flash solidifies Xiaomi's ambition to be a key architect in that future, challenging the current hierarchy in the global AI landscape directly from its base in Beijing.
