Frontier AI models now cost as much as $5 per million input tokens and $30 per million output tokens, roughly 25 times the $0.20 and $1.25 charged for an earlier lightweight model, according to new research from McKinsey's QuantumBlack unit on the economics of agentic AI published in August 2026. Even as individual token prices fall industry wide, enterprise spending on AI is spiking as companies deploy more agents across more workflows.

That pricing gap is reshaping how companies budget for AI agents. McKinsey found that customer facing AI agents in banking can cost $20,000 to $30,000 to run as a single agent workflow, and $100,000 to $200,000 for a multiagent team, once fixed infrastructure and orchestration costs are included alongside token spending.

Human review, more than token consumption, drives the bulk of ongoing costs. In banking customer service tasks, McKinsey found token costs represent just 20 to 25 percent of variable run costs, while human oversight accounts for 70 to 75 percent, since 10 to 20 percent of agentic runs in areas such as customer onboarding are checked by risk and compliance staff.

Scale changes the math considerably. Using a conversational agent to onboard 2,500 new bank customers a year costs $10,000 to $15,000, McKinsey found, and doubling that volume raises costs only to $15,000 to $20,000, since fixed costs spread across more transactions. Applied to a full onboarding workflow, McKinsey estimates the total cost per customer can fall from $50 to $150 down to $10 to $30 as agent design matures.

Source: McKinsey QuantumBlack - https://www.mckinsey.com/capabilities/quantumblack/our-insights/where-ai-agents-pay-off-a-practical-guide-to-the-economics-of-agentic-workflows