Xiaomi has quietly entered the global AI race with its MiMo-V2-Pro, a trillion-parameter model that briefly masqueraded online as an unreleased DeepSeek model.

The model excels in coding, agentic tasks, and creative storytelling, outperforming rivals like Claude on price. It scored 78% on SWE-bench Verified and ranked eighth globally on Artificial Analysis.

Its sibling, MiMo-V2-Omni, offers true multimodal capabilities. Meanwhile, MiMo-V2-TTS focuses on text-to-speech.
In creative writing tests, MiMo-V2-Pro delivered rich, culturally nuanced narratives surpassing even Claude Opus 4.6.

However, it struggles with advanced mathematics and occasionally falters in logical reasoning transparency.

Despite limitations, its competitive pricing-$1 per million input tokens, $3 per million output-makes it attractive for developers.
