Long-horizon agent
An agent that must continuously complete multi-step decisions and environmental interactions over a relatively long time span. It typically operates as a loop of planning, calling tools, observing results and re-planning, and a single task may trigger dozens of model calls.
Timeline (last 90 days)
A long-horizon agent refers to an agent that needs to continuously complete multi-step decisions and environment interactions over a relatively long time span, typically manifesting as a loop of planning, calling tools, observing results, and planning again, with a single task potentially triggering dozens of model calls.
Competition in long-horizon agents will ultimately come down to who can stably complete more complex tasks with fewer tokens.
All Facts (2)
A long-horizon agent refers to an agent that needs to continuously complete multi-step decisions and environment interactions over a relatively long time span, typically manifesting as a loop of planning, calling tools, observing results, and planning again, with a single task potentially triggering dozens of model calls.
50%UnverifiedCompetition in long-horizon agents will ultimately come down to who can stably complete more complex tasks with fewer tokens.
50%