What are Claude usage credits?

People type this because the word “credits” gets attached to three completely different things in Claude, and the pages that currently come up answer one and skip the other two. Here is the short version, then each one untangled, current as of June 2026.

Matthew Diakonov, Written with AI

Published June 17, 20266 min read

Direct answer · verified 2026-06-17

Claude usage credits are a prepaid, pay-as-you-go balance you add to a paid plan (Pro, Max 5x, or Max 20x) so you can keep working after you hit your plan’s included limits. Spending past the limit draws down that balance at standard API pricing rates, charged separately from the subscription. Your base plan itself is not a credit balance, it is metered in tokens over a rolling window. And the separate “Agent SDK credit” announced for June 15, 2026 was paused on the day it was due to launch, so there is no such credit to claim right now.

Source: Anthropic Help Center, “Usage credits for paid Claude plans”.

Three different things hide under “credits”

Almost every confusion about Claude credits comes from collapsing these into one. They bill differently, they reset differently, and two of them are not credits at all. Keep them separate and the whole topic gets simple.

1. Included plan limits (not credits)

What you already pay for on Pro or Max. Metered in tokens over a rolling 5-hour session window plus weekly caps. It is a rate limit that resets, not a dollar balance. When you run out, you wait for the window to reset.

2. Usage credits (the real answer)

A prepaid dollar balance you opt into with 'Add funds'. Once your included limit is gone, further usage bills against this balance at standard API rates, separately from your subscription. Available on Pro, Max 5x, and Max 20x.

3. The Agent SDK credit (paused)

Announced May 2026 to move Agent SDK, claude -p, and third-party app usage onto a separate monthly pool ($20 / $100 / $200) from June 15. Anthropic paused it on June 15 itself. Those surfaces still draw from your normal subscription limits.

What actually happens when you hit your limit

This is the moment the word “credits” starts to matter. The path forks depending on whether you have a prepaid balance.

From included limit to paid usage credits

You burn through the included token limit

Weighted by context size, model, and tool calls. Long-running windows hit this faster because the full history re-sends every turn.

No prepaid balance? You are blocked until reset

With usage credits at zero, Claude simply stops until the rolling window or weekly cap resets. No surprise charge.

Funds added? Overflow bills at standard API rates

If you have prepaid usage credits, consumption past the limit draws down that balance at per-token API pricing, separately from your subscription.

Balance hits zero, you are blocked again

Usage credits are prepaid, not a line of credit. When the balance empties you add more funds or wait for the included window to reset.

Quick test: is the thing in front of you actually a credit?

When you see “credit” in the Claude UI, a billing email, or a third-party blog post, run it against this. Real usage credits satisfy all three.

A real Claude usage credit is...

A dollar balance you prepay, not a token rate limit that resets
Opt-in via 'Add funds', so leaving it at zero means no overage charge
Spent at standard API pricing rates, charged separately from your plan
The same thing older posts call 'extra usage' (Anthropic renamed the wrapper, not the mechanism)
A separate monthly 'Agent SDK credit' you can claim today
Included in your Pro or Max subscription as a fixed credit bundle

Why so many recent posts are already wrong

A wave of articles in late May and early June 2026 told people to “claim a free $20 credit” or warned that agent and claude -p usage would leave their subscription on June 15. That referred to the Agent SDK credit: a planned split that would have given each plan its own monthly pool ($20 for Pro, $100 for Max 5x, $200 for Max 20x) and moved automated SDK usage off the subscription.

Anthropic paused that change on June 15, 2026, the same day it was scheduled to take effect, saying it is reworking the plan to better support how people build with Claude subscriptions and will give advance notice before any future change. So if a guide tells you to go claim an Agent SDK credit, it is describing a thing that does not currently exist. The only credit you can actually add money to today is the prepaid usage credit balance covered above.

Net effect as of June 17, 2026: Agent SDK, the headless claude -p command, and third-party apps still draw from your normal Pro, Max, Team, and Enterprise limits. There is no separate pool to claim, and your subscription limits are unchanged.

Where this meets Fazm, and the lever you actually control

Fazm is a native macOS app that wraps the real Claude Code agent loop over ACP. It signs in with your existing Claude Pro or Max account, so everything on this page applies unchanged: the included token limits, the optional prepaid usage credits, the same shared pool across claude.ai, Desktop, and Claude Code. There is no Fazm metering layer and no markup. If you prefer, the custom API endpoint field lets you point the same loop at a proxy or gateway, and the billing follows that endpoint instead.

The number you can actually influence is how fast you burn the included limit before you ever touch a paid credit. Because usage is weighted by how much context re-sends every turn, a window you keep open for hours costs more per turn than a fresh one. Most tools quietly auto-compact long sessions to fight this, which trims tokens but silently drops decisions you made earlier. Fazm does not auto-compact. Instead it gives every chat a one-click fork: when the accumulated context gets heavy, you fork to a fresh window from a clean checkpoint, dropping the token weight without losing the thread you care about. That fork is the difference between a long session that slowly drains your included limit and one that holds a clean, cheaper context for the work that matters.

The practical takeaway

Usage credits are the overflow valve. Context discipline is the tap. Control the tap with forking instead of compaction, and you reach for the valve far less often.

Run Claude Code on your own plan, without the context tax

Walk through how Fazm keeps long sessions cheap with one-click forking and no auto-compacting, all on your existing Pro or Max account.

Frequently asked

Frequently asked questions

What exactly are Claude usage credits?

They are a prepaid balance you add to a paid Claude plan (Pro, Max 5x, or Max 20x) so you can keep working after you hit your plan's included limits. Anthropic's help center calls them 'usage credits'. When the balance is active, anything beyond your included limit is billed against that prepaid balance at standard API pricing rates, charged separately from your subscription. You top it up by clicking 'Add funds' and choosing an amount.

Are usage credits the same as my Pro or Max plan limit?

No, and this is the part most people get wrong. Your included plan usage is metered in tokens over a rolling 5-hour session window plus weekly caps. That is not a credit balance, it is a rate limit that resets. Usage credits are the optional overflow on top of that: a dollar balance you prepay and only spend once the included token limits are exhausted.

Did Anthropic add a separate Agent SDK credit on June 15, 2026?

No. Anthropic announced in May 2026 that Agent SDK, claude -p, and third-party app usage would move to a separate monthly credit starting June 15, 2026 ($20 for Pro, $100 for Max 5x, $200 for Max 20x). On June 15 it paused that change on the day it was due to take effect. Those surfaces still draw from your existing Pro, Max, Team, and Enterprise subscription limits, there is no credit to claim, and your subscription limits are unchanged.

How are usage credits billed, as a flat fee or pay-as-you-go?

Pay-as-you-go at standard API pricing rates. You prepay a balance, and consumption past your plan limit draws down that balance at the same per-token rates Anthropic charges API customers. It is not a flat add-on or a fixed monthly bundle of credits.

Does Fazm add its own credits or markup on top of Claude?

No. Fazm runs the real Claude Code agent loop and signs in with your existing Claude Pro or Max account, so usage hits your own plan and the same shared limits described here. There is no separate Fazm metering and no markup on tokens. If you would rather route through an API key, a corporate proxy, or any Anthropic-compatible gateway, Fazm has a custom API endpoint field for that, and the billing follows that endpoint instead.

What burns through my included limit fastest, before I ever touch paid credits?

Context size. Usage is weighted by conversation length, model choice, and tool calls, and every turn re-sends the live context. A long-running window costs more per turn than a short one even with identical prompts. Large CLAUDE.md files, codebase search, MCP tool definitions, and retries all add weight. The slower you burn the included token limit, the later you ever dip into a prepaid usage credit balance.

Can I turn usage credits off so I never get charged overage?

Yes. Usage credits are opt-in. If you do not add funds, hitting your included limit simply blocks further use until the window resets, with no surprise charge. Adding funds is what enables the pay-as-you-go overflow; leaving the balance at zero keeps you on the included-limits-only behavior.