In a significant move within the competitive AI landscape, Chinese tech giant Xiaomi has officially launched and open-sourced its latest large language model, MiMo-V2-Flash. The model, which was unveiled at Xiaomi's "Human-Car-Home Full Ecosystem" Partners Conference on December 17, 2025, promises a compelling combination of high performance, remarkable speed, and aggressive pricing, positioning itself as a new contender among open-source AI models.
Xiaomi Unveils MiMo-V2-Flash with Focus on Speed and Efficiency
The core promise of the MiMo-V2-Flash model is its efficiency. Designed from the ground up with inference speed in mind, early user feedback highlighted its surprisingly fast response times compared to other popular models like DeepSeek, Doubao, and Yuanbao. This speed is not just a user experience perk but is architecturally central to the model's design, which Xiaomi claims is key to enabling high-bandwidth agent collaboration. The model utilizes a Mixture of Experts (MoE) architecture with a total of 309 billion parameters, of which 15 billion are activated for any given task, allowing it to maintain a large knowledge base while optimizing computational efficiency.
Model Specifications:
- Name: Xiaomi MiMo-V2-Flash
- Architecture: Mixture of Experts (MoE)
- Total Parameters: 309 Billion (309B)
- Activated Parameters: 15 Billion (15B)
- Key Focus Areas: Reasoning, Code Generation, AI Agent Applications
Performance and Pricing Aim to Disrupt the Market
Xiaomi is making bold claims about MiMo-V2-Flash's capabilities, particularly in code generation and tool use, which are essential for AI agents. The company states that its code ability rivals that of the leading closed-source model Claude Sonnet 4.5. Perhaps more disruptive is its pricing strategy. The API is priced at CNY 0.7 per million input tokens and CNY 2.1 per million output tokens. Xiaomi emphasizes that this makes MiMo-V2-Flash's inference cost merely 2.5% of Claude Sonnet 4.5's, presenting a potentially game-changing value proposition for developers and enterprises. The model weights and inference code have been fully open-sourced under the permissive MIT license, and a limited-time free API tier is available alongside a live web demo.
Pricing & Availability:
- API Pricing: Input: CNY 0.7 / Million Tokens; Output: CNY 2.1 / Million Tokens
- Cost Claim: Inference price is 2.5% of Claude Sonnet 4.5.
- Licensing: Model weights and code open-sourced under MIT License.
- Availability: Limited-time free API tier available; Web Demo is live.
The "AI Prodigy" Behind the Model Takes Center Stage
The launch event also served as the official debut for Luo Fuli, the newly appointed head of the MiMo project at Xiaomi. The 95-year-old technical expert, often referred to as an "AI prodigy" in Chinese tech circles, joined Xiaomi in November after high-profile stints at Alibaba's DAMO Academy and DeepSeek's parent company, Fantasia Quant. Her recruitment, reportedly pursued personally by Xiaomi founder Lei Jun, was a topic of significant industry speculation. At the conference, Luo outlined the model's three key technical characteristics: superior code and tool-calling capabilities for agent communication, an architecture designed for reasoning efficiency, and a new post-training paradigm for stable reinforcement learning scaling.
Performance Claims:
- Speed: User-reported faster response times vs. DeepSeek, Doubao, Yuanbao.
- Code Ability: Positioned as comparable to Claude Sonnet 4.5.
- Ranking: Reported to be "top two" in code and agent capabilities on world-class evaluation benchmarks.
Xiaomi's Broader AI Ambitions Come into Focus
The release of MiMo-V2-Flash is not an isolated event but a key milestone in Xiaomi's accelerated AI strategy. During the conference, Xiaomi Group President Lu Weibing revealed a broader portfolio of MiMo models, including a 7B reasoning model, vision-language models (MiMo-VL), an audio generation model (MiMo-Audio), an on-device VLM, and an embodied AI model. Lu stated that Xiaomi has initiated "pressure-style investment" in AI, with progress "far exceeding expectations," and will focus its future efforts on the "deep integration of AI with the physical world." The launch of this high-performance, cost-effective model provides a foundational tool for that ambitious vision, potentially powering smarter devices across Xiaomi's ecosystem of phones, cars, and home products.