Xiaomi MiMo-V2-Pro (mimo.xiaomi.com)

Xiaomi has announced the MiMo-V2-Pro, its latest flagship foundation model designed for advanced agentic workloads, capable of orchestrating complex workflows in real-world applications. Ranking 8th globally in comprehensive intelligence and 2nd among Chinese LLMs, this model boasts over one trillion parameters, with 42 billion active. It features an efficient hybrid attention mechanism that enhances inference while supporting a remarkable 1M-token context, making it well-suited for diverse productivity scenarios beyond simple Q&A tasks. Significantly, MiMo-V2-Pro surpasses competitors in coding capabilities and multi-step reasoning, achieving superior results in major benchmarks, including coding tasks where it competes effectively with models like Claude 4.6 Opus. It is particularly optimized for deep integration with agent frameworks such as OpenClaw and is available as an API with tiered pricing. Xiaomi emphasizes that this model will be a vital step toward achieving artificial general intelligence (AGI), with ongoing research aimed at enhancing its reasoning and decision-making capabilities in complex, real-world environments.