Why did our AI bill go up even though model prices came down?

The per-token sticker price did fall, but the moment an enterprise plan flips to pay as you go, the same work runs five to ten times more, because you are metered on consumption rather than on seats. Both facts are true at once, which is why the chart and the invoice disagree.

The Disconnect Between Sticker Prices and Consumption Billing

Vendors highlight declining per-token rates as cost savings, and those sticker prices have indeed fallen. However, when an enterprise moves from seat-based subscriptions to consumption billing, identical workloads can cost five to ten times more because organizations are billed on metered token usage rather than predictable seats.

Source: Your AI Got Cheaper. Your Bill Didn’t.

Subscription Versus Metered Token Cost Realities

A flat developer subscription costs roughly $200 each month with unlimited usage. Switching that same developer doing the exact same work to per-token metering can increase costs to $1,200 a month or more—a sixfold increase for the identical output.

Source: Build or Buy: The Arithmetic on Running Your Own Model

Incentives Tied to Token Volume

The biggest labs are monetising hard on consumption billing: Anthropic passed OpenAI in business spend almost entirely because of Claude Code, and the labs make more money when you burn more tokens.