Anthropic changes pricing structure - again

For the third time this year

Anthropic has walked back a recent announced pricing change that would have significantly raised costs for power users. But how long can the company afford to subsidise usage?

In the middle of last month, Anthropic announced it would treat the Claude Agent SDK - including via third-party apps and the “claude -p” command - separately from standard Claude use. From the 15^th June, usage was set to be billed at Anthropic’s API rates.

Previously, Anthropic only limited Agent SDK use according to the standard weekly caps on a user’s Claude subscription tier. Power users utilising AI agents could leverage this to get a substantial discount versus paying API fees, as Anthropic was effectively subsidising agent use by 15-30x compared to API pricing.

By billing Claude Agent SDK at API rates Anthropic was set to close some of its funding gap, but also deliver a “major cost increase” for heavy users.

However, the company has now updated its billing support page to say it is not – at this stage – moving ahead with that plan.

“We're pausing the changes to Claude Agent SDK usage described below. For now, nothing has changed: Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription's usage limits. The previously announced monthly credit, which would have been available to eligible claimants in connection with these changes, isn't available.

“We’re working to update the plan to better support how users build with Claude subscriptions. When we have an update, we'll share it before anything takes effect.”

The AI industry doesn’t know what to do with pricing

Token-based billing, aka consumption-based pricing, is the tech industry’s newest hot topic, though AI developers wish it wasn’t.

Companies including OpenAI, Github and Anthropic have realised they can’t afford to effectively pay users to adopt their tools by subsidising the cost of LLM queries. Instead, each has recently changed their terms of use to charge based on token usage.

Seeing as there is little to no clue how many tokens a given prompt or agent will consume, and poorly worded prompts may require multiple iterations to reach the desired result, this has led to opaque billing cycles and climbing costs.

Some companies have even moved away from their existing partnerships altogether.

With an IPO coming up, Anthropic is keen to retain as many users as possible – which might be why it’s hit reverse on its earlier decision. But how long can it keep subsidising costs?