Abstract

We define Agency as the emergent capacity of AI systems to function asautonomous agents actively discovering problems, formulating hypotheses, andexecuting solutions through self-directed engagement with environments andtools. This fundamental capability marks the dawn of the Age of AI Agency,driven by a critical industry shift: the urgent need for AI systems that don'tjust think, but work. While current AI excels at reasoning and generatingresponses, industries demand autonomous agents that can execute tasks, operatetools, and drive real-world outcomes. As agentic intelligence becomes thedefining characteristic separating cognitive systems from productive workers,efficiently cultivating machine autonomy becomes paramount. Current approachesassume that more data yields better agency, following traditional scaling lawsfrom language modeling. We fundamentally challenge this paradigm. LIMI (Less IsMore for Intelligent Agency) demonstrates that agency follows radicallydifferent development principles. Through strategic focus on collaborativesoftware development and scientific research workflows, we show thatsophisticated agentic intelligence can emerge from minimal but strategicallycurated demonstrations of autonomous behavior. Using only 78 carefully designedtraining samples, LIMI achieves 73.5% on comprehensive agency benchmarks,dramatically outperforming state-of-the-art models: Kimi-K2-Instruct (24.1%),DeepSeek-V3.1 (11.9%), Qwen3-235B-A22B-Instruct (27.5%), and GLM-4.5 (45.1%).Most strikingly, LIMI demonstrates 53.7% improvement over models trained on10,000 samples-achieving superior agentic intelligence with 128 times fewersamples. Our findings establish the Agency Efficiency Principle: machineautonomy emerges not from data abundance but from strategic curation ofhigh-quality agentic demonstrations.