Cursor Says Agent Token Costs Fell 7% Without Quality Loss

Cursor said it has cut token costs in its coding agent by 7% without reducing agent quality, attributing the savings to tighter prompts, selective tool loading, better caching and compressed file reads. The change is a cost-efficiency update rather than a new model launch, but it speaks directly to one of the central operating pressures in AI coding tools: keeping long-running agent workflows affordable while preserving reliability.