LMAS removes waiting cost, not wake-up cost.

The completion prompt still costs tokens. The saving comes from eliminating the polling loop while the long-running job is still running.

Observed baseline

In one OpenCode loop, repeated status checks cost $2.72 over 10.2 minutes. Projected over a 24-hour wait, that is about $382 and 1.93B tokens.

Observed LMAS lifecycle

The measured LMAS lifecycle includes the start handoff and completion handling. It does not grow with wall-clock wait time. In the measured run, handoff plus completion cost about $0.18 and 387K tokens.

What this means

• If the job runs for one minute, the savings are small.
• If the job runs overnight, the avoided polling dominates.
• If the model is expensive or context is large, the avoided context reloads matter even more.