Xiaomi Unveils MiMo-V2-Flash, Its New Open-Weight AI Model to Take on DeepSeek

Why It Matters

MiMo‑V2‑Flash gives Xiaomi a foundational AI asset that can power its hardware ecosystem while challenging established model providers on cost and performance.

Key Takeaways

• 309B‑parameter model with 150 tps inference speed
• Costs $0.1M input, $0.3M output tokens
• Open‑weight, downloadable via MiMo Studio and Hugging Face
• MoE architecture balances performance and efficiency
• Benchmarks match DeepSeek V3.2, surpass Kimi K2 long‑context

Pulse Analysis

The AI landscape is increasingly dominated by large‑scale models that power everything from chat assistants to code generators. Xiaomi’s entry with MiMo‑V2‑Flash signals a strategic shift from pure hardware manufacturing to owning a foundational model stack. By releasing the model under an open‑weight license and distributing it through popular developer hubs, Xiaomi lowers the barrier for developers to experiment, fostering a community that can accelerate innovation around its ecosystem of smartphones, tablets, and electric vehicles.

Technically, MiMo‑V2‑Flash leverages a Mixture‑of‑Experts (MoE) design coupled with a hybrid sliding‑window attention (SWA) mechanism, allowing the 309‑billion‑parameter network to scale efficiently. The architecture reduces the need to re‑process long prompts, cutting inference costs to $0.1 per million input tokens and $0.3 per million output tokens—prices that undercut many commercial offerings. Benchmarks show the model achieving 73.4% on SWE‑Bench Verified and matching Anthropic’s Claude 4.5 on coding tasks, while also excelling in long‑context reasoning against competitors like Moonshot’s Kimi K2.

From a market perspective, MiMo‑V2‑Flash positions Xiaomi as a credible rival to DeepSeek, Anthropic, and OpenAI, especially as the company integrates AI agents directly into its consumer devices. This vertical integration could create a feedback loop: richer AI capabilities enhance device appeal, driving higher adoption rates that, in turn, generate more data to refine the model. For investors and industry observers, Xiaomi’s AI push underscores the growing importance of proprietary models in differentiating hardware brands and could reshape competitive dynamics in the Chinese and global AI markets.