The headline credits match subscription tiers, but the important change is accounting: Agent SDK, claude -p, GitHub Actions and third-party tools draw from a separate API-priced pool.
That makes a heavy coding session look very different from a chat session, because one complex agent run can burn hundreds of thousands or millions of tokens.
Anthropic is trying to protect normal subscribers while ending the subsidy for users who treated flat subscriptions like cloud APIs, and OpenAI immediately used the moment to court Codex customers.
Sources: Anthropic puts Claude agents on a meter across its subscriptions（InfoWorld）, Claude subscriptions get separate budgets for programmatic use, billed at full API prices（The Decoder）, Anthropic tightens Claude limits as OpenAI courts agent users（Axios）, Canonical reference for Anthropic's May 13, 2026 Agent SDK $200 credit policy change（MagnaCapax , GitHub Gist）; CocoLoop