HomeFeaturesPricingBlogFAQContact
← All articles

Anthropic Paused Its Agent SDK Billing Change on June 15 -- Here Is What Was Planned and What Comes Next

Key takeaways
  • Anthropic announced on May 14, 2026 that Agent SDK, claude -p headless, Claude Code GitHub Actions, and third-party ACP apps would move from subscription pools to a separate metered credit starting June 15 -- then paused the change on that same date.
  • The planned credit amounts were $20/month for Pro, $100/month for Max 5x, and $200/month for Max 20x, all billed at standard API list rates with no rollover and no pooling across team members.
  • Two Claude 4 model IDs -- claude-opus-4-20250514 and claude-sonnet-4-20250514 -- were retired as scheduled on June 15; requests to those model strings now fail with no redirect or fallback.
  • Anthropic says it is "reworking the plan to better support how users build with Claude subscriptions" and will give advance notice before any revised billing split launches.

June 15, 2026 was supposed to be a significant date for Claude subscribers. Anthropic had announced -- exactly one month earlier -- a structural change to how Agent SDK usage would be billed, alongside the scheduled retirement of two older Claude 4 model IDs. One of those two things happened as planned. The other did not.

What Anthropic Had Announced

On May 14, 2026, Anthropic published a billing change that would take effect on June 15. The core shift: usage through the Claude Agent SDK, the claude -p headless command, Claude Code GitHub Actions, and any third-party app built on Anthropic's Agent Client Protocol (ACP) would be removed from existing subscription limits and billed through a separate monthly credit pool.

The credit amounts by plan were:

Those credits would be billed at standard API list rates -- for example, $3 per million input tokens and $15 per million output tokens for Claude Sonnet 4.6. Credits were per-user, not pooled across teams. They would not roll over. Once exhausted, automated requests would stop entirely unless the user had manually opted into overflow billing at their own API rate.

The reasoning behind the change, though Anthropic did not state it explicitly, is legible from the numbers. Analysis published in June by developer tooling research site MorphLLM estimated that heavy Agent SDK users were extracting 25 to 36 times the compute value implied by their monthly subscription fee. A $20 Pro subscription was, in practice, delivering hundreds of dollars of agentic inference in active months. That is a subsidy that does not scale.

Why It Was Paused

On June 15 itself, Anthropic confirmed in its Help Center and in subscriber notifications that the billing split would not launch as scheduled. The statement: the company is "reworking the plan to better support how users build with Claude subscriptions" and will provide advance notice before any revised version launches.

As of now, nothing has changed for subscribers. Agent SDK usage continues to draw from existing subscription limits as before. There are no new credit pools to claim, no billing impact, and no deadline.

What likely complicated the timeline: Anthropic simultaneously lost access to its two flagship models the previous week due to the US export control directive affecting Claude Fable 5 and Mythos 5. Rolling out a billing overhaul while the company's most capable models were offline -- and while thousands of developers were scrambling to adjust workflows -- was a reasonable place to press pause.

What Did Happen on June 15 -- Model Retirements

The model retirement timeline ran as announced. Two Claude 4 model IDs were deprecated on June 15 and now return errors:

These are not soft deprecations. Requests to the old model strings fail, not degrade and not silently redirect. Any production code pinned to those specific model IDs needed to be updated before June 15. If your application was using them and you have not yet migrated, fix the model string first.

What the Pause Signals

The underlying economics that drove Anthropic's billing plan have not changed; only the launch date has. Flat-rate subscriptions and agentic token consumption are pulling in opposite directions. A single complex agentic coding task can push 400,000 to 2 million tokens through the API. At Claude Sonnet 4.6 rates, one demanding session can cost more than a Pro user's entire monthly subscription fee.

Earlier this month, reporting surfaced that Microsoft had cancelled internal Claude Code licenses for some engineering teams after token bills reached approximately $2,000 per engineer per month -- compared to GitHub Copilot Enterprise's $39 flat seat fee. That math is why Anthropic was trying to restructure billing in the first place.

The change is paused, not cancelled. A revised plan is coming. BYOK users who connect API keys directly -- as tools like ByteChat allow -- already operate on API-rate billing and are unaffected by however Anthropic eventually restructures its subscription tiers.

Frequently asked questions

Did Anthropic's June 15 billing change go into effect?

No. Anthropic announced on June 15, 2026 that it was pausing the Agent SDK credit split before it launched. The company said it is "reworking the plan" and will give advance notice before any revised billing change goes live. Subscription usage continues as before.

Which Claude models were retired on June 15, 2026?

Two Claude 4 model IDs -- claude-opus-4-20250514 and claude-sonnet-4-20250514 -- were retired as scheduled. Requests to those model strings now fail. The replacements are claude-opus-4-8 and claude-sonnet-4-6 respectively.

How would metered Agent SDK credits have compared to direct API access?

Under the planned (paused) change, Pro users would have received $20/month in Agent SDK credits billed at standard API rates -- roughly equivalent to about 6.7 million input tokens on Sonnet 4.6. BYOK users who connect their own API keys pay those same rates but have no monthly credit cap; they pay for what they use and nothing stops automated requests when a notional credit pool runs out.

API users are immune to subscription billing overhauls

ByteChat connects your own API keys directly to Claude, GPT, Gemini, and more -- with zero markup. No subscription credit pools, no surprise billing caps, no waiting for Anthropic to "rework the plan."

Try ByteChat free →