At the opening of the 2026 AI Industry Conference on Oct 15, the China Academy of Information and Communications Technology (CAICT) released the 'Top 10 AI Industry Keywords 2026,' outlining — from technology, application and ecosystem layers — a complete picture of AI's accelerated diffusion this year.
Model self-evolution describes how developers increasingly use frontier models to speed up AI's own R&D; current practice remains 'bounded self-evolution' within human-set loops. Agent engineering is coalescing around a path where 'models provide reasoning, harnesses organize and constrain tasks.' Agent terminals fuse agents with devices — agent phones, AI glasses, home hubs — moving from single-agent to multi-agent swarm intelligence.
Embodied real-world training is shifting from 'demo mode' to 'work mode,' advancing along three directions: scaling data, matching scenarios and deploying pilots. Chip-model synergy deepens coordination on operator performance, interconnect and memory scheduling — a key to scaling domestic compute. Other keywords include high-value AI application scenarios, where AI's penetration of the real economy moves from 'whether there is a scenario' to 'which scenarios are worth doing.'
The list signals China's AI industry maturing from headline model benchmarks toward engineering, productization and deployment in physical and industrial settings.
