Skip to content
Loop Engineering
Loop Engineering Tools

Loop Engineering Cost Calculator

Estimate per-run, daily, monthly, and worst-case token cost for an agent loop — then get optimization tips before you schedule it. Pricing presets are examples only.

Pricing presets are examples. Verify current provider pricing before making budget decisions.
Est. monthly cost
$58.03
Risk level
Medium
Per iteration
$0.0465
Per run
$0.4836
Daily cost
$1.93
Monthly cost
$58.03
Worst-case / mo
$66.96
Model
Example balanced model
Optimization suggestions
  • This loop looks budget-aware. Keep the stop rule, retry limit, and human approval gate in place before scheduling it.
Suggested runtime caps

Turn this estimate into real stop conditions. These map to continuous-claude flags ($1.00 per-run cap ≈ 1.5× the per-run estimate). Verify before use.

caps.sh
# continuous-claude caps derived from this estimate — verify before use
--max-runs 8
--max-cost 1
--max-duration 1h

How Loop Costs Are Calculated

Cost is per-iteration token cost multiplied by iterations, retries, runs per day, and days. The worst-case figure assumes a retry multiplier of at least 1.5 so you can size a budget cap for a bad day.

cost-model
costPerIteration = (inTokens / 1e6) * inputPrice
              + (outTokens / 1e6) * outputPrice
costPerRun   = costPerIteration * (maxIterations * retryMultiplier)
dailyCost    = costPerRun * runsPerDay
monthlyCost  = dailyCost * 30
worstCase    = costPerIteration * maxIterations * runsPerDay * 30
              * max(retryMultiplier, 1.5)

Token-Rich vs Token-Poor Loop Engineering

Token-rich

Token-rich loops pass large context into every iteration. They are easier to implement but more expensive and more likely to hit context limits.

Token-poor

Token-poor loops pass only summaries, recent observations, or retrieved memories. They are cheaper and faster but require better memory and state management.

Loop Budget Optimization Tips

  • If input tokens are large, switch to summary or retrieval memory and pass only the last few relevant observations.
  • If iterations are high, add a hard stop condition and require a report after repeated failures.
  • If the retry multiplier is high, improve validation and fallback before scheduling the loop.
  • If monthly cost is high, require human approval before long-running loops and add budget caps.