Cursor cuts agent token costs by 7% with no drop in quality

Your move: A behind-the-scenes cost cut, nothing to change on your end

Cursor published a blog post on September 23, 2026 describing changes that cut its agents' token costs by about 7% with no drop in quality. The savings came from a shorter system prompt, loading tools only when a task needs them instead of listing every tool up front, better placement of prompt caching, and removing instructions that older models needed but current models don't. Cursor said cold-cache rates, when saved context can't be reused, dropped by about 20%.

Source: Cursor Blog ↗

Published by Agent Log. The summary reports the linked source; “Why it matters” and the verdict are Agent Log's interpretation.

More Cursor updates →