Xiaomi launches an AI model to power advanced AI agents and coding tasks

Xiaomi has launched a new artificial intelligence model called MiMo-V2-Pro, expanding the company’s efforts beyond hardware and into advanced AI systems. The model is designed to support intelligent software agents capable of performing complex tasks rather than only generating text responses. The announcement marks Xiaomi’s growing focus on building AI technologies that can power real-world applications.

MiMo-V2-Pro designed as a foundation model for AI agents

MiMo-V2-Pro is described as Xiaomi’s flagship foundation model intended to act as the core intelligence behind agent-based systems. Instead of functioning only as a conversational AI, the model is designed to coordinate workflows, manage software development tasks and assist in engineering-related operations. The brand’s goal is to move beyond simple question answering toward systems that can complete multi-step tasks.

ALSO READ: OpenAI launches GPT-5.4 Mini and Nano to power faster high-volume AI workloads

The model is built with more than one trillion total parameters, with 42 billion active at any given time. It also supports a context window of up to one million tokens, which allows the system to process long documents and extended conversations without losing context. Xiaomi says this design helps the model handle complicated workflows that require long sequences of instructions or information.

Benchmark results released by the company indicate that MiMo-V2-Pro performs competitively with several well-known AI models across agent and coding related tests. Its performance is reported to be close to leading models like ChatGPT (GPT-5.2) and Claude Opus 4.6 in several coding and agent benchmarks, though it generally still trails them slightly in overall top-tier evaluations.

Developer tools and API access now available

Xiaomi has also opened API access for developers who want to experiment with the new model. The API supports the full 1M-token context window and offers different pricing tiers depending on the amount of context used during processing.

According to Xiaomi, MiMo-V2-Pro was previously tested under the internal name “Hunter Alpha” on the OpenRouter platform, where it reportedly handled large volumes of requests during early trials. The company says feedback from developers during this testing phase helped improve long-context stability and agent task performance.

The MiMo team states that future work will focus on improving long-horizon reasoning and decision making, with the aim of building systems capable of solving increasingly complex real-world problems.

ALSO READ: Google Maps gets chatty with new Ask Maps feature for smarter local searches

Stay tuned to Unboxed by Croma for all the updates from the world of artificial intelligence.