Anthropic shifts pricing
Anthropic moved enterprise Claude pricing from simple subscriptions to usage-based billing, signaling firms will now pay more directly for how much AI they consume. A report says the change shifts billing to a base fee plus additional usage charges for heavy users, and Anthropic’s pricing docs reflect the new structure. Users have complained about recent performance hiccups and a lack of transparency, prompting public backlash alongside the pricing change. ( )
Anthropic has shifted Claude Enterprise from mostly flat subscriptions to a seat fee plus metered usage, pushing heavy corporate users toward pay-as-you-go bills. (claude.com, theinformation.com) Anthropic’s pricing page now lists a self-serve Enterprise plan at $20 per seat per month when billed annually, with usage billed separately at application programming interface rates based on what a team uses. Its sales-assisted Enterprise tier says custom terms, usage commitments and invoicing are available, with a minimum of 50 seats. (claude.com) The company’s broader pricing page still shows Team at $25 per seat per month billed annually, while Enterprise is described as a “tailored approach” with security features, usage analytics and spend controls. Anthropic also says prices and plans can change at its discretion. (claude.com, claude.com) The change lands as Anthropic has been tightening how customers use Claude. Axios reported on April 6 that Anthropic blocked Claude subscriptions from powering third-party agent tools such as OpenClaw, a move that pointed users toward more directly metered access. (axios.com) The Information reported on April 14 that businesses with heavy Claude usage are likely to pay significantly more under the new model, and tied the shift to a broader compute crunch across the artificial intelligence industry. That is the core business problem here: training and running large models requires expensive graphics processing units and cloud capacity, and flat plans are harder to sustain when customers run autonomous coding agents for hours at a time. (theinformation.com, axios.com) At the same time, Anthropic is dealing with complaints from developers who say Claude’s coding performance has slipped and its limits are harder to understand. Fortune reported on April 14 that users accused the company of a decline in quality and of failing to explain changes clearly. (fortune.com, aol.com) Anthropic executives have pushed back on some of those claims. VentureBeat reported that the company had been tuning settings across user segments, while Boris Cherny, who leads Claude Code, said Anthropic changed the default effort level to “medium” after feedback that Claude was using too many tokens. (venturebeat.com, github.com) That response has not ended the dispute, because many users say lower effort settings, changing cache behavior and percentage-based usage bars make costs harder to predict. Anthropic’s own status page says it tracks real-time and historical system performance, but the current pricing pages give enterprise buyers only the basic structure: seats, metered usage and spend controls. (status.claude.com, claude.com) For companies standardizing on Claude, the immediate result is simpler to state than to budget: access now starts with a per-seat fee, but the final bill depends on how much artificial intelligence work employees and agents actually run. (claude.com, theinformation.com)
