Anthropic Cancels June 15 Claude Agent SDK Billing Change

Anthropic emailed Claude subscribers on June 16, 2026 to call off a planned June 15 billing change that would have moved Claude Agent SDK, claude -p, and third-party app usage off subscription plans and onto a separate, metered credit pool. Subscription limits stay exactly as they were. There’s no credit to claim and nothing for users to switch on.

The note went out by email only. Anthropic posted no blog, no status update, and no developer changelog. For the heavy automation users who spent a month rewriting scripts and pricing out alternatives, the reversal arrived as a single line in their inbox.

Anthropic Pulls the June 15 Split With a One-Line Email

The message was short. “We’re not making this change today,” Anthropic told subscribers, referring to the credit split it had announced on May 14. The wording matters. “Today” is not “never,” and the company gave no new timeline, no replacement plan, and no explanation for the about-face.

What stays in place is the status quo developers already knew. Automated jobs running through the Agent SDK or headless claude -p keep drawing from the same usage allowance as interactive chat. Pro, Max 5x, and Max 20x plans bill the way they did in May.

The quiet rollout is the tell here. A company confident in a pricing decision announces it on its newsroom and stands behind it. A company emailing a reversal at the last minute, with no public post and no press line, is managing a problem it would rather not discuss in the open.

What the Scrapped Billing Change Would Have Done

The cancelled plan would have carved programmatic usage out of subscriptions entirely. Under the May 14 announcement, Agent SDK calls, claude -p, Claude Code GitHub Actions, and any third-party app authenticating through a Claude plan would have stopped counting toward plan limits. Instead they’d draw from a fixed monthly dollar credit, metered at full standard API list prices with no rollover.

Each tier got a flat allowance. Once it ran dry, automated work either stopped or billed at API rates that dwarf a flat subscription. Anthropic’s own Claude plan pricing page lists the same Pro, Max 5x, and Max 20x tiers that the credit pool would have repriced for anyone running agents at scale.

For light users the credits looked generous. For heavy ones the math turned ugly. One widely shared developer analysis pegged the effective price increase at 12x to 175x depending on workload volume, since power users routinely burn through far more compute than a flat $200 buys at list rates.

Interactive use was never in scope. Claude.ai web chat, the terminal version of Claude’s model lineup including the new Fable 5 release, and Claude Cowork would have kept drawing from subscription limits. The split was aimed squarely at automation, the exact workload that grew fastest over the past year.

Why a Class Action Landed the Same Week

The reversal didn’t happen in a vacuum. Days earlier, a Washington subscriber named Karl Kahn filed a proposed class action against Anthropic in the US District Court for the Northern District of California, accusing the company of overselling its Max tiers. The suit targets the $100 Max 5x and $200 Max 20x plans and seeks refunds for buyers going back to the tiers’ launch in April 2025.

The complaint argues the advertised multipliers never held up. It alleges the “Max 20x plan delivers just six to eight times the usage of Pro, while Max 5x delivers just three-and-a-half times the usage of Pro,” far below the headline numbers. Kahn says one five-hour coding session ate 15% of his weekly quota. Anthropic added those weekly caps in late August 2025 and said at the time they’d affect fewer than 5% of subscribers, a figure that now sits at the center of the dispute. Rival pricing pressure isn’t helping either, as covered in our report on OpenAI weighing token price cuts as Anthropic gains ground.

This is a really good type of case for a class action, because the law is very clear that companies have to be honest about their advertising and marketing.

That assessment came from Kati Daffan, founding partner at Vaca Daffan LLP, the firm representing Kahn. The suit seeks refunds and changes to how Anthropic markets its plans, which makes any further billing shift legally awkward while the case is live.

The Third Retreat in Six Months

This is the third time in six months Anthropic has pushed a usage-policy change and then walked it back or recast it. The pattern says the company still hasn’t settled on how to bill the developers who run Claude hardest. Each attempt has drawn the same response from the people it targets.

It started in winter. The sequence reads like a company testing the edge of what its base will tolerate.

1. January 9, 2026: Anthropic activated server-side blocks at 2:20 AM UTC that stopped subscription OAuth tokens from working in third-party tools, with no notice, then reversed within days after backlash.
2. February 2026: The company revised its Consumer Terms of Service to formally bar using plan-based OAuth tokens outside Claude Code and Claude.ai.
3. May 14, 2026: Anthropic announced the Agent SDK credit split set for June 15.
4. June 14, 2026: Kahn’s class action hit the federal docket.
5. June 16, 2026: Anthropic emailed subscribers that the split was off.

The OAuth fight in January cut deep because it broke tools developers had built businesses on. OpenCode, with more than 126,000 GitHub stars, and OpenClaw both lost subscription access overnight. Ruby on Rails creator David Heinemeier Hansson was among the prominent critics.

The technical boundary Anthropic drew then still stands in its Claude Code overview docs. Subscription tokens are for first-party tools. Everything else routes through the paid API. The June plan was the billing half of that same fence, and it met the same wall of resistance.

Anthropic is also fighting on other fronts, from competitive pricing wars to geopolitics, including the fallout we covered when Macron and Modi pitched cooperative AI after US curbs on Anthropic. A pricing climbdown plus a class action plus regulatory noise is a lot of pressure landing in one month.

What Heavy Claude Users Should Do Now

Nothing changes for now, and that’s the practical headline. Automations that ran on a subscription on June 14 keep running on it today. There’s no migration to perform, no credit to redeem, and no new cap to budget around.

• Keep running Agent SDK and claude -p jobs on your existing plan, since they still count toward normal limits.
• Hold off on rebuilding workflows around a credit pool that no longer exists.
• Watch your inbox, not the blog, because Anthropic delivered this reversal by email with no public post, per Anthropic’s Claude Agent SDK documentation staying unchanged.
• Stay inside first-party tools if you rely on OAuth, since the third-party restriction remains in force.

Frequently Asked Questions

Do I need to do anything now that Anthropic canceled the billing change?

No. The June 15 credit split never took effect, so there’s nothing to switch on, migrate, or claim. Your Pro, Max 5x, or Max 20x plan bills exactly as it did before. Anthropic confirmed the cancellation by email on June 16, 2026, and said only that it’s “not making this change today.” Keep an eye on future emails for any new timeline.

Will my Claude Agent SDK automations still run on my subscription?

Yes. Agent SDK and headless claude -p jobs keep drawing from your plan’s normal usage limits, the same as they did in May. The plan to route them through a separate $20 to $200 metered credit pool is shelved. If a script worked on June 14, it works today with no code or auth changes needed.

Could Anthropic bring back the credit pool later?

Very likely. The word “today” in Anthropic’s email signals a pause, not a permanent decision, and the company has reworked usage policy three times since January 2026. No replacement date exists yet. Heavy programmatic users should plan for some future metered model and keep API-rate cost estimates handy in case the split returns in another form.

Does this reversal end the class action lawsuit?

No. Karl Kahn’s class action over Max 5x and Max 20x usage limits remains active in the Northern District of California and is separate from the cancelled billing change. The suit covers conduct since the tiers launched in April 2025 and seeks refunds. Track the docket through the federal court system or follow filings from Vaca Daffan LLP for updates on class certification.

Can I still use Claude with third-party tools like OpenCode?

Not through your subscription. Anthropic’s February 2026 Terms of Service still bar plan-based OAuth tokens in any tool outside Claude Code and Claude.ai. That restriction did not get reversed. To use Claude inside third-party tools legitimately, you need a separate paid API key billed at standard list rates, not your Pro or Max login.

For all the noise, Anthropic ended the month roughly where it started, with the same subscriptions and the same unanswered question about how it will eventually charge its heaviest automation users. Developer pressure stopped the split, and a lawsuit raised the cost of pushing it through. The next attempt, whenever it lands, will arrive under far closer watch.