Xiaomi Group announced that its MiMo-V2.5 series of large AI models has officially entered public beta testing. According to industry research firm Artificial Analysis, the flagship MiMo-V2.5-Pro model ranks first among global open-source large models and is within the top five overall worldwide.
The MiMo-V2.5 series includes the flagship MiMo-V2.5-Pro, the multimodal MiMo-V2.5, as well as MiMo-TTS and MiMo-ASR models. Xiaomi stated that MiMo-V2.5-Pro excels at handling complex multi-step agent tasks and demonstrated the ability to complete a graduate-level compiler project in 4.3 hours that typically requires weeks to finish.
Benchmark tests show MiMo-V2.5-Pro performing on par with top closed-source models such as GPT-5.4 and Claude Opus 4.6, while offering significant cost efficiencies. Xiaomi has also updated its Token Plan billing system, removing long-context premiums and introducing discounts for night usage, auto-renewals, and annual plans.
