Anthropic has ended the flat-rate enterprise seat plan it used to sell — the one that gave customers a predictable monthly ceiling on what they would pay for Claude. Now the ceiling is gone. In its place is a base fee of $20 per user per month plus usage-based charges: every query, every code completion, every background task accrues to the bill. The shift, which took full effect in early March, is a repackaging of the same licensing move Oracle made in the 2000s when it converted perpetual database contracts to per-core and per-user metrics, betting that the cost of staying would always exceed the cost of the increase. Anthropic is making the same bet with Claude. (The Register)
The transition was gradual. Anthropic began renewing enterprise customers under usage-based terms in November 2025, according to a company spokesperson. In February it introduced a single $20 per employee monthly tier covering access to Claude, Claude Code, and Cowork. By March 8, internal documentation shows the legacy Standard and Premium seat plans were no longer available to new contracts, with existing customers migrated at their next renewal date. The old plans carried flat fees of up to $200 per user per month with token pools that scaled with team size. (The Register)
For some enterprise customers, the shift is cosmetic. One AI consultancy, IntuitionLabs, told The Register that for many of its clients the base seat fee represented only about 20 percent of the total bill — the other 80 percent was already metered API usage on top of the flat rate. For those customers, the new model is less a price increase than a clarification of what they were already paying. But for customers whose teams were heavy but not overflow users, the flat fee acted as a de facto cap. That cap is gone. (The Register)
The timing is not neutral. Anthropic's revenue run rate reportedly reached $30 billion in early 2026, up from $9 billion at the end of 2025 — a trajectory that required enterprise adoption at scale. Killing the legacy plan before an IPO lets Anthropic exit a pricing structure that would complicate how investors model recurring revenue, converting flat-rate commitments into a usage-based model that scales with consumption rather than against a contractual ceiling. (Implicator.ai)
There is also the infrastructure question. Anthropic's API uptime over the 90 days ending April 8 was 98.95 percent — below the 99.99 percent standard established cloud providers commit to. Retool founder David Hsu publicly explained his decision to move his company to OpenAI not on price but on reliability: Anthropic's model was better, he told the Wall Street Journal, but the service kept dying. Under the new pricing model, every failed retry, every cache refresh, every subagent loop that burns tokens without producing output is a line item on the bill. Reliability and cost are now inseparable. (Implicator.ai)
Anthropic declined to comment on the specific terms of customer contracts or the IPO timeline. The company's own support documentation, updated in February, frames the new model plainly: the seat fee covers access to the platform, not usage. Every token consumed by chat, code, or collaboration is billed separately at standard API rates. (Anthropic support documentation)
For enterprise buyers, the practical question is not whether Claude is worth the price but whether the price is predictable. A CFO who budgeted $100,000 a year for 500 seats at the flat rate now faces a model where that budget is a floor, not a ceiling. The next renewal is the audit.
