Anthropic puts Claude agents on a meter across its subscriptions
Pricing change separates long-running agent usage from the flat-rate seat — the first explicit acknowledgment from a major lab that agents and unlimited subscriptions don't add up.
Anthropic has shifted Claude agents to a metered pricing model across its subscription tiers, decoupling long-running agentic compute from the flat-rate seat that powers most consumer and prosumer plans.
The change is small in implementation, large in implication. Interactive chat usage stays inside the seat. Agent usage — particularly anything that runs unattended across many tool calls — gets metered against a separate counter that bills above the subscription baseline.
Anthropic has not publicly disclosed exact per-unit pricing in subscription-tier documentation, and the precise structure differs across Pro, Max, and Team. The shape, though, is clear: the flat seat covers chat; agents cost what agents cost.
The forcing function is the unit economics. A typical chat session terminates when the user closes the tab. An agent session does not. It runs in the background, makes tool calls the user never sees, and consumes meaningfully more inference than a chat session — sometimes by an order of magnitude.
Expect every other frontier lab on a flat-rate plan to follow within two quarters. The math is the math.