Xiaomi’s MiMo-V3 to adopt new architecture as HySparse2 cuts long-context costs

Xiaomi MiMo lead Fuli Luo said MiMo-V3 will adopt a new architecture. Its core technology, HySparse2, is designed to reduce the cost of long-context inference and improve retrieval for agentic workloads. At a context length of 1 million tokens, HySparse2 reduces prefill computation by 5.02 times and cuts the KV cache by 4.5 times.