AI + ML

Anthropic squeezes enterprises by ejecting bundled tokens from seat deal

Large organizations pushed toward metered pricing

UPDATED More bad news for Claude users. Anthropic has revised its seat-based pricing for enterprise customers, shifting them to a new pricing plan upon contract renewal.

The transition began in November 2025, according to a company spokesperson, when Anthropic began renewing companies under a usage-based plan. In February, the company introduced a single $20-per-employee monthly fee that covered various Claude services – Code, Cowork, chat – on a usage-based plan.

At some point after that (no later than March 8), company documentation changed to say that the legacy enterprise plan would no longer be available.

Current support documentation says, “Chat-only seats and Standard/Premium seats are no longer available for new contracts – both legacy plan types are transitioning to the single Enterprise seat at their next renewal.”

The change comes at a time of turmoil for the company, which in the run-up to a rumored IPO has seen demand for Claude AI services outpace capacity, resulting in adjustments to terms of service and rate limits designed to ensure availability.

Anthropic offers three individual subscription tiers (Free, Pro at $20 per month, and Max at $100 or $200 per month), a Team tier for small organizations ($25 per month or $125 per month), and an Enterprise tier for large organizations (negotiated).

Individual and seat-based subscriptions impose time and usage limits but keep the price flat, with extra usage beyond those limits on a pay-as-you-go basis.

Anthropic also sells pay-as-you-go API access, metered at varying token rates, which the company refers to as "consumption billing." 

Buying Claude through the API has been significantly more expensive than buying it through subscription plans, an ongoing source of friction with customers who resent recent policy changes that have disallowed subscription plans with third-party tools.

What's changed is that the seat-based component – the flat monthly fee and associated token pool – for Enterprise no longer comes with subsidized tokens. The company's legacy documentation used to say, "Some Enterprise organizations are on seat-based plans with per-seat usage limits rather than consumption billing ... Seat-based plans charge a flat monthly fee per seat and include usage limits."

IntuitionLabs, an AI consultancy for the pharmaceutical industry, previously published an analysis of Anthropic's pricing options. CEO Adrien Laurent told The Register in an email that the restructured pricing only applied to enterprise customers.

"In our experience at IntuitionLabs, our Enterprise clients were already spending significantly on overage," said Laurent. "Most users were in extra usage territory anyway, well beyond what the base seat fee covered.

"What's actually changing is that the seat fee no longer bundles any usage allowance at all. Every token gets billed at standard API rates on top of the base seat. For some of our clients, the base seat was only about 20 percent of their total bill, and the other 80 percent was already metered API usage. For them, the shift won't feel dramatic. For lighter users who stayed inside the bundle, it will."

Asked whether AI pricing needs to be more predictable, Laurent said it already is, if you pay published API rates.

"What hasn't been predictable are the 'buffet' subscription packages," he said. "The rules around those seem to change every other week. Session caps, weekly quotas, what counts as usage, which third-party tools are allowed. That's where the unpredictability lives."

The flat-rate plans, Laurent said, effectively subsidize usage.

"It's well known that the $200 Max plan gives you several thousand dollars worth of API-rate credits if you actually push it," he said. "Unfortunately, enterprises don't have access to Max. It's a consumer plan. But even at that level of effective consumption per seat, the business value is clearly there. Most users are enjoying a real discount."

Laurent expects AI tooling and models to improve in ways that reduce token expenditures, through better compression, smarter document conversion, and improved routing.

"That should relieve some of the pressure," he said. "But it's also very possible that Anthropic is acquiring customers faster than it can scale capacity, and the unit economics simply don't work at the old prices. The subsidy has to come down somewhere."

The danger, Laurent argues, is that Anthropic may lose interest in serving small organizations and individuals if enterprise spending gets large enough. "At that point, they could move toward a billing model that only works for enterprises, and the flat-rate consumer and prosumer plans that got the whole ecosystem here could quietly go away," he said. ®

Correction: An earlier story mistakenly gave the impression that the changes are brand new, when in fact they occurred in February.