Claude usage credits: two separate meters, and which one your tools actually spend from

Usage credits are how paid Claude plans (Pro, Max 5x, Max 20x) keep working past their included usage limits. There are two distinct systems. The first is interactive extra-usage credits: prepaid funds you enable so that when you hit your plan cap while chatting or running Claude Code in the terminal, your next requests bill at standard API rates instead of cutting you off. The second, effective June 15, 2026, is a separate monthly programmatic credit pool that only specific programmatic surfaces draw from. They are charged separately from your subscription and show up as additional line items on your bill.

Anthropic grants $20/month to Pro, $100/month to Max 5x, and $200/month to Max 20x. These figures are corroborated across Anthropic's announcement coverage. Team and Enterprise have their own per-seat allowances. The programmatic credit is consumed at Anthropic's public API rates (per million input and output tokens), the same as pay-as-you-go API billing.

Programmatic usage that draws from the new credit pool: the Claude Agent SDK (Python and TypeScript), the claude -p one-shot CLI command, Claude Code GitHub Actions, and third-party apps built on the Agent SDK. Interactive usage that still comes out of your normal plan limits: chatting with Claude on web, desktop, or mobile, using Claude Code in the terminal, and Claude Cowork. The dividing line is roughly automation versus a human at the keyboard.

No. The monthly programmatic credit resets with each billing cycle and does not carry over, so unused credit is lost at the end of the month. Interactive extra-usage credits are prepaid funds you add yourself; those persist as a balance until spent, but they only kick in after you exceed your plan's included usage.

It depends entirely on how the app authenticates. If the tool logs in with your Claude subscription via OAuth (the same login the terminal uses) and runs the actual Claude Code agent loop, its usage is billed like interactive Claude Code: against your plan's normal limits, not the programmatic pool. If instead the tool uses an ANTHROPIC_API_KEY or the Agent SDK, that usage is API-metered and would draw from the programmatic credit (or your extra-usage balance) at API rates. The auth mechanism, not the UI, decides the bucket.

fazm runs the real Claude Code agent loop through ACP and supports three billing modes, chosen at startup in acp-bridge/src/index.ts. Mode B (the default for people who bring their own Claude account) launches with your OAuth subscription token and no API key, so usage draws from your plan's normal limits like the terminal does. Mode A uses a provided API key, which is API-metered. Mode C routes to a custom Anthropic-compatible endpoint. You pick which meter you spend from by which mode you run.

In Claude, go to Settings then Usage, find the Usage credits section, and click Enable. You then prepay by clicking Add funds and purchasing an amount. After that, once you exceed your plan's included usage, subsequent requests bill against that prepaid balance at standard API rates rather than stopping. The interactive daily redemption limit is capped (Anthropic lists a per-day ceiling in the same settings page).

Before the change, subscription plans allowed near-unlimited programmatic access, so automated agents and pipelines could consume far more compute than an interactive subscriber. Anthropic announced on May 14, 2026 that starting June 15, 2026 programmatic usage would draw from a dedicated monthly credit instead, putting heavy automated workloads on metered API economics while leaving normal chat and terminal coding on the flat subscription.