Cost per taskdefinition and how it works in 2026
- Cost per task
- The fully-loaded cost of an AI completing one unit of work — model spend + infrastructure + integration cost amortized + retries. The right unit for AI ROI math.
Cost-per-task is the metric that makes AI ROI legible. Per-token costs are too granular (a single task can use 10K–500K tokens depending on workflow); per-seat costs are too coarse (one seat might handle 100 or 100,000 tasks per month). Cost-per-task — "how much does it cost the agent to handle one ticket, generate one PR, book one meeting" — is the metric that maps to business value.
The honest math includes: model inference cost (per-token × tokens used) + infrastructure (serving, observability, vector DB amortized) + integration cost (one-time, amortized over expected lifetime) + retries + failed attempts. Vendor pricing pages quote the first; mature buyers compute all of it.
Typical 2026 cost-per-task ranges: tier-1 support resolution $0.50–$3.50, AI SDR qualified meeting $50–$200, ai-coding-agent PR generation $5–$30, document summarization $0.02–$0.50, deep-research query $0.50–$10. Compare against the human-handled cost-per-task to know whether the agent pencils.
Frequently asked
How is cost-per-task different from per-task-pricing?+
Per-task-pricing is the vendor's billing model (you pay $X per task). Cost-per-task is your fully-loaded cost (vendor billing + your infrastructure + your integration + your overhead). They're often different by 30–80%.
What's a good cost-per-task target?+
It should be 30–60% of the equivalent human-handled cost-per-task. Higher (>60%) and the ROI is too thin to justify deployment overhead. Lower (<30%) and you're probably under-counting hidden costs and will be surprised at year 2 renewal.