As usage of Anthropic’s coding-focused AI tools has surged, the company has tweaked its pricing model in a way that could make it significantly more expensive for some enterprise users.
The Information reported on Tuesday that Anthropic has shifted Claude Enterprise subscriptions away from a flat fee of up to $200 per user per month, which included a set amount of discounted tokens, to a usage-based model that charges for computing use on top of a $20 monthly seat fee per user.
Claude Enterprise is Anthropic’s business-focused bundle, which includes tools like Claude Code and Claude Cowork. These products are particularly compute-intensive, often running for extended periods of time to complete complex tasks. However, as adoption of these tools has grown, so have the costs required to run them, putting pressure on Anthropic’s margins. The Information reported that weekly active users of Claude Code doubled between January and February.
Fredrik Filipsson, co-founder of Redress Compliance, which helps companies negotiate software licensing agreements, told the outlet the changes could potentially triple costs for some enterprise customers.
Following the publication of this story, an Anthropic spokesperson told Gizmodo that the pricing change was first introduced in November 2025. The Information noted that some customers only began noticing the shift in recent weeks.
A review of Anthropic’s Enterprise support page using the Wayback Machine shows the change was reflected by January of this year.
“Usage-based Enterprise plans charge a base seat fee plus actual usage costs drawn from a shared organizational pool,” the page said. At the time, it listed two seat types: Chat and Chat + Claude Code.
By February, the page had been updated to state that all new Enterprise customers would be on “the usage-based billing model,” while customers who were on legacy plans would still retain the seat-based model.
In the weeks that followed, Anthropic made additional changes to how it charges for specific tools. The spokesperson told Gizmodo that, in February, the company introduced a single all-inclusive Enterprise seat covering chat, Claude Code, and Cowork for $20 per user.
The most recent version of the support page states that legacy plans are “transitioning to the single Enterprise seat at their next renewal.” And the legacy plans are dead.
“The structure is built for how enterprise customers actually deploy Claude,” the spokesperson told Gizmodo in an emailed statement. “Every seat includes full access to chat, Claude Code, and Cowork, so there’s no friction deciding who gets what at purchase. Usage is pooled across the org, so light users don’t create waste and heavy users don’t hit walls.”
The changes come amid growing complaints online that Anthropic may have made recent tweaks to its models that worsened their performance.
One particular complaint, posted to GitHub in February by a senior director at AMD, has since gone viral across social media.
In the post, Stella Laurenzo wrote that Claude Code could no longer be trusted for complex engineering work. She said the model appeared to decline in performance in February compared to January, including ignoring instructions and providing “simplest fixes” that were incorrect.
Users on X began sharing screenshots of the post this month, with one writing, “basically: anthropic sneakily turned down how hard claude thinks before editing code, changed the default from “high” to “medium” effort, and hid the reasoning from session logs. all without telling users.”
Claude Code creator Boris Cherny responded on X, calling the allegation “false.”
“We defaulted to medium as a result of user feedback about Claude using too many tokens. When we made the change, we (1) included it in the changelog and (2) showed a dialog when you opened Claude Code so you could choose to opt out. Literally nothing sneaky about it — this was us addressing user feedback in an obvious and explicit way.” Cherny wrote.
These are just the latest examples of how AI companies will need to tweak both their products and pricing strategies as demand for their models explodes, sending costs through the roof, and stakeholders insist on seeing a path to profitability after unprecedented levels of investment.
Correction 4/17: A previous version of this article suggested that Anthropic’s usage-based pricing model was new as of April. The usage-based pricing model was introduced late last year, and The Information notes that many enterprise customers have only begun to notice the change as their legacy plans expire.
