The Multi-Agent Workflow Cost per Task Calculator estimates the unit cost of a completed coordinated workflow. It combines per-agent model usage, supervisor or coordination tokens, paid tool charges, shared infrastructure, and full-workflow reruns.
Use it to evaluate whether multi-agent quality gains justify their added expense, compare different agent counts, or set budget limits for automated operations. Input and output token prices are separated, and coordination costs are shown explicitly rather than hidden inside one average.
1. Set workflow volume
Enter the number of completed workflows in the cost period.
2. Describe agent usage
Enter agent count and average input and output tokens used by each agent.
3. Add orchestration tokens
Include supervisor, synthesis, routing, and shared-state token usage.
4. Enter current prices and charges
Provide model prices plus tool and infrastructure cost per workflow.
5. Apply rerun overhead
Estimate full-workflow reruns and review unit, model, and total cost.
The result is the estimated variable cost of one successfully completed multi-agent workflow after rerun overhead.
This model prices coordination tokens at the input rate. If coordination uses output generation or a different model, calculate that portion separately and enter an equivalent cost.
Given: 2,000 workflows, 3 agents, 2,000 input and 500 output tokens per agent, 1,200 coordination tokens, $2 input and $8 output per million tokens, $0.025 tools, $0.01 infrastructure, and 5% reruns.
Calculation: Per-agent model cost = (2,000 ÷ 1,000,000 × $2) + (500 ÷ 1,000,000 × $8) = $0.008. Three agents cost $0.024. Coordination = 1,200 ÷ 1,000,000 × $2 = $0.0024. Base = $0.024 + $0.0024 + $0.025 + $0.01 = $0.0614. Adjusted = $0.0614 × 1.05 = $0.06447.
Result: Cost is about $0.0645 per workflow and $128.94 for 2,000 workflows.
Calculate each agent role separately and add their model costs, then enter the combined amount through an adjusted tool or infrastructure field.
Yes, when the same context is sent independently to every agent. Include it in input tokens per agent.
The calculator uses the input-token price as a simplification; adjust externally when coordination uses another billing category.
Their expected cost is represented through the rerun rate, expressed as full-workflow equivalents.
Reduce duplicated context, route easy tasks to smaller models, limit review rounds, cache shared state, and avoid paid tools when they add little value.