Xiaomi MiMo V2.5 multimodal AI launch boosts efficiency and agentic coding

Xiaomi has launched MiMo V2.5 and MiMo V2.5 Pro, aiming to combine multimodal processing and agentic tool use in a single workflow (image, audio, video). The headline update is MiMo V2.5, which supports multimodal inputs without switching models, making “see-hear-act” capabilities more streamlined. Xiaomi claims MiMo V2.5 Pro is near top-tier for coding agents, with 57.2% task resolution on SWE-bench Pro (vs. a typical ~25% average). It is also designed for long-horizon work targeting 1,000+ tool calls, and supports a 1M-token context window. Speed is reported at 60–80 tokens/second for the Pro tier. Pricing and efficiency are central to the rollout. MiMo V2.5 Pro is listed at $1.00 per million input tokens and $3.00 per million output tokens. Xiaomi also says MiMo V2.5 Pro uses 42% fewer tokens than Kimi K2.6 for comparable results, while the base MiMo V2.5 consumes nearly half the tokens versus similar systems. During launch, Xiaomi removed extra charges tied to full 1M-context usage and reset user credits. For everyday users, MiMo V2.5 targets 100–150 tokens/second with $0.40 per million input tokens and $2.00 per million output tokens, while still offering the 1M-token context window. For crypto traders, this is primarily an AI-tech-sector signal rather than a direct on-chain catalyst; any market impact would likely be indirect via future demand for “AI x crypto” tooling and infrastructure. MiMo V2.5 remains the key product point to watch as Xiaomi pushes efficiency and agentic coding workflows at clearer unit economics.