Anthropic Enterprise Edition is switching to a usage-based pricing model, with heavy users potentially seeing their costs double or triple.

According to 1M AI News monitoring, in recent weeks, Anthropic has switched Claude Enterprise from a fixed subscription model to a usage-based billing model. Previously, enterprise customers had a maximum fixed fee of $200/person/month, which included a certain amount of discount token usage. Now, it has been changed to a base seat fee of $20/person per month plus an additional fee based on actual computational power consumption. Fredrik Filipsson, co-founder of software licensing consultancy Redress Compliance, estimates that costs for some heavy users will double to triple. Companies with fewer than 150 people are not affected, and the Team subscription (up to $100/person/month) remains unchanged.

Anthropic stated that under the old model, some customers frequently hit the usage limit, causing work interruptions, while others did not fully utilize the prepaid quota. The new model "better reflects the actual situation of the workload shifting from fixed seats to agent-based usage." The driving factor is the significant increase in usage of Claude Code and the AI work assistant Claude Cowork. Agent-based products can run continuously for long periods or even execute automatically on a schedule, with computational power consumption far exceeding that of traditional conversational scenarios. Sudip Roy, former Head of Inference at Cohere and co-founder of Adaption Labs, pointed out that subscription models are essentially priced based on the assumption that users will not fully utilize the quota; once agents consume the full quota, "the profit margin is almost completely eroded."

This is not a unique issue to Anthropic. Replit and Cursor adjusted their pricing last summer due to cost pressures from agent-based products, and when Salesforce launched Agentforce at the end of 2024, it directly adopted a usage-based billing model. When AI agents can consume computational power uninterrupted, the underlying assumption of seat-based pricing is no longer valid. In contrast, OpenAI took a different approach: in early April, they introduced a new $100/person/month tier for Codex (with quotas between the $20 Basic tier and the $200 Pro tier), attempting to attract teams that require cost predictability with a fixed price point.