by PENG Peng

Xiaomi, a Chinese smartphone and consumer electronics maker, on Thursday unveiled three in-house large models — Xiaomi MiMo-V2-Pro, Xiaomi MiMo-V2-Omni and Xiaomi MiMo-V2-TTS — stepping up its push into the global AI agent race.

The company said APIs for MiMo-V2-Pro and MiMo-V2-Omni are now available, enabling developers to build and deploy agent-based applications.

MiMo-V2-Pro, the flagship text model, is designed for complex tasks such as reasoning, planning and tool use. MiMo-V2-Omni integrates text, vision and audio for end-to-end execution, while MiMo-V2-TTS focuses on speech synthesis to deliver more natural voice interaction.

The model supports context windows of up to 1 million tokens, longer than most mainstream models, enabling more complex workflows.

Pricing is a key differentiator. Xiaomi said MiMo-V2-Pro is priced below comparable offerings, starting at US$1 per million input tokens and US$3 per million output tokens for contexts up to 256,000 tokens, rising to US$2 and US$6 for 1 million-token contexts. MiMo-V2-Omni is priced at US$0.4 per million input tokens and US$2 per million output tokens.

The pricing underscores intensifying competition in China's AI sector, where companies are using cost advantages to attract developers and expand market share.

Xiaomi is offering one week of free API access in partnership with agent framework teams including OpenClaw, OpenCode, KiloCode, Blackbox and Cline, aiming to quickly build a developer ecosystem.

Founder LEI Jun said the company's AI-related research and capital spending will exceed 16 billion yuan (US$2.3 billion) this year.

According to Xiaomi, MiMo-V2-Pro ranks eighth globally on the Artificial Analysis benchmark and fifth among model providers. Lei said the model would be upgraded rapidly.

In December, LU Weibing, partner and president of Xiaomi Group, announced the open-sourcing of Xiaomi MiMo-V2-Flash, saying the company had begun "high-intensity investment" in AI and that progress in large models and applications had "far exceeded expectations."