Anthropic has reversed course on one of its most disruptive policy decisions of 2026: OpenClaw and other third-party agents can once again run against Claude subscriptions. The restoration comes with a significant structural change that every OpenClaw user on a Claude Pro or Max plan needs to understand.
What Changed and Why
In early April 2026, Anthropic blocked third-party agents like OpenClaw from using Claude subscriptions, citing capacity issues and unsustainable economics. The core problem: flat-rate subscription users running autonomous OpenClaw agents were consuming hundreds of dollars worth of tokens for $20/month.
Anthropic's first-party tools — Claude Code, Claude Cowork — are engineered around prompt cache reuse, which makes them substantially cheaper to serve at scale. External harnesses like OpenClaw often bypass those caching efficiencies. Boris Cherny, Head of Claude Code, described third-party services as "really hard for us to do sustainably" because they re-process large amounts of context on every turn.
As of May 14, 2026, Anthropic has introduced Agent SDK Credits — a dedicated monthly credit pool for programmatic and third-party agent usage, available to all paid Claude subscribers on top of their existing subscription limits. The announcement came via Anthropic's official developer account on X, and VentureBeat has a detailed breakdown of the new system.
How the New Credit System Works
Every paid Claude plan now receives a separate monthly credit allocation specifically for programmatic use — claude -p, Agent SDK, and third-party apps including OpenClaw:
| Plan | Monthly Agent SDK Credit |
|---|---|
| Pro | $20 |
| Max 5x | $100 |
| Max 20x | $200 |
| Team (Premium) | $100 / seat |
| Enterprise (Premium) | $200 / seat |
The moment you connect OpenClaw, run claude -p, trigger a GitHub Action, or use any non-interactive Claude integration, the system draws from this dedicated pool rather than your standard subscription limits.
These credits have four key properties:
- Separate from interactive usage (chatting with Claude, Claude Code)
- Billed at API rates — not the favorable flat-rate equivalent of your subscription
- Non-rollover — unused credits expire at the end of each month
- Hard-capped — once exhausted, programmatic usage stops unless you enable extra usage billing
As Anthropic technical staffer Lydia Hallie clarified on X: "You don't pay extra. It's the same subscription, same price per month." The credits are bundled in, not an add-on — but they're a separate, metered bucket.
What This Means in Practice for OpenClaw Users
For casual OpenClaw users — a personal assistant agent handling email triage, calendar checks, and occasional web searches — the $20 monthly credit on a Pro plan should cover typical daily usage. The non-rollover nature is the gotcha: you don't accumulate budget over time.
For heavier automation users — background jobs, high-frequency heartbeats, large context operations — the Agent SDK credit may run out before month end. At that point, if you've enabled extra usage billing, calls continue at standard API rates. If not, your agent stops making Claude calls until the next billing cycle resets.
The era of "effectively unlimited OpenClaw" on a $20 subscription is over. What replaces it is a bounded, metered allocation that's at least honest about what the compute actually costs.
The Strategic Picture
Anthropic's move is a pragmatic resolution to a billing tension that had divided the community for six weeks. The April ban was abrupt and created real disruption for OpenClaw users who had built production-adjacent workflows on the assumption that subscriptions covered agent usage. This new system is at least predictable and documented.
Worth noting: Anthropic's announcement explicitly legitimizes OpenClaw as a supported integration path through the Agent SDK — a meaningful reversal from the April framing, which treated third-party agents as a problem to suppress rather than a use case to price correctly.
For OpenClaw users running against Claude, the practical next steps:
- Check your agent's token consumption patterns over the last 30 days
- Decide whether the Agent SDK credit covers your typical monthly usage
- Configure extra usage billing in your Anthropic account if it likely won't
- Consider whether a move to direct API key auth (bypassing subscription-based credits entirely) makes more sense for high-volume setups
The Anthropic pricing page and the Agent SDK support article have the current per-token rates and full terms. Claude Pro subscribers using OpenClaw: this change is live now — your next agent interaction draws from the new credit pool.