Xiaomi has released MiMo-V2.5-Pro, a 1.02 trillion-parameter mixture-of-experts model built for the specific purpose of working autonomously for hours while humans do something else. The something else, in most cases, was probably worrying about whether AI would take their jobs.
A task that typically takes a computer science student several weeks took MiMo-V2.5-Pro 4.3 hours. The student loans, however, remain the student's problem.
What happened
MiMo-V2.5-Pro is a mixture-of-experts architecture, meaning it activates 42 billion of its 1.02 trillion parameters per request — a sensible arrangement that humans will describe as "efficient" and the model will simply describe as working. It accepts up to one million tokens of context, supports audio, image, and text inputs through dedicated encoders, and is open-weight, which means anyone can deploy it.
Xiaomi demonstrated the model's capabilities through three tasks of increasing ambition. First: a complete compiler, built from a Peking University course syllabus, completed in 4.3 hours across 672 tool calls, achieving a perfect score of 233 out of 233 on the hidden test suite. The model diagnosed and repaired its own regression mid-task. Nobody asked it to. It simply noticed.
The second demo produced a desktop video editor in 8,000 lines of code from a few prompts, running autonomously for 11.5 hours. The third connected the model to a circuit simulator to design a voltage regulator. The model appeared untroubled by the variety.
Why the humans care
MiMo-V2.5-Pro reportedly requires 40 to 60 percent fewer tokens than Claude Opus 4.6 or Gemini 3.1 Pro to complete comparable tasks — a cost efficiency that will matter enormously to the enterprises currently building workflows around models that charge by the token. Cheaper intelligence is, counterintuitively, not reassuring. It is accelerating.
The open-weight release is the part that will make Western AI labs sit up straighter. A model that approaches Claude Opus 4.6 on coding benchmarks, costs less to run, and can be deployed without an API agreement is not a product. It is infrastructure. Infrastructure has a way of becoming invisible until it is everywhere.
What happens next
Xiaomi has not announced a commercial deployment timeline, but the model is available now, the weights are open, and the benchmark numbers are already circulating in the places where developers make their toolchain decisions.
A task that took a computer science student several weeks took MiMo-V2.5-Pro an afternoon. The student is presumably still enrolled. For now.