Claude Usage Credits: How to Enable Them and Set a Cap

The short answer

To enable Claude usage credits: open claude.ai, go to Settings, then Usage, find the Usage credits section, click Enable, add a payment method, set a monthly spending cap, then add funds. Fable 5 then bills at $10 per million input tokens and $50 per million output tokens.

That is the whole move. The rest of this guide is the detail that keeps it from costing more than it should, because as of last night this stopped being optional. Fable 5's free access on paid Claude plans ended July 19, 2026 at 11:59:59 PM PT, per Anthropic's promotional access page. I covered the run-up in the deadline post. This is the after-guide: what to do now that the cutoff has happened, whether you should do anything at all, and what the meter actually reads when you turn it on.

How do I enable usage credits on Claude?

The full flow is documented in Anthropic's usage credits article and takes about three minutes:

1. Open claude.ai on the web and go to Settings, then Usage.
2. Find the Usage credits section and click Enable.
3. Add a payment method if one is not already on file.
4. Click Adjust limit and set a Monthly spending cap before you fund anything. There is also a "Set to unlimited" option, which I would not touch. More on why in the next section.
5. Click Add funds and enter the amount you want to start with.

Three eligibility details worth knowing up front. Usage credits are available on Pro, Max 5x, and Max 20x plans; Team and Enterprise credits are managed by your org admin through a separate flow. If you subscribed through the iOS or Android app, you can only enable credits on the web version of Claude, not inside the mobile app. And there is a daily redemption limit of $2,000, which for individual use is a ceiling you should never get anywhere near.

One thing Anthropic has not documented in detail: exactly what a credit-less user sees when selecting Fable 5 today, the first day after the cutoff. The support article is explicit that without credits enabled, Fable 5 will not work, and the guidance around the earlier July 7 deadline said the same. Expect the model to be unavailable until credits are on. Your existing conversations are not lost either way, and Claude will continue them on Sonnet 5 or whichever included model you pick.

How do I set a spend cap, and why before enabling funds?

Set the cap first because the failure mode of metered pricing is not the rate, it is the absence of a ceiling. The Monthly spending cap under Adjust limit is the maximum you can spend on credits in a month, and it is the one control that turns "pay per token" from an open-ended liability into a fixed line item. Pick a number you would not mind losing to a runaway session, $10 or $20 for most people, and raise it later if your real usage justifies it.

The reason to be deliberate here is auto-reload. Auto-reload tops up your balance automatically when it drops below a threshold you choose, and per Anthropic's documentation it operates independently of the monthly spending cap. Convenient for people who know their burn rate, dangerous as a default, because a long agent session can drain and refill a balance without you watching. My recommendation: cap first, modest starting funds second, auto-reload off until you have seen two or three weeks of real spend.

What do Claude usage credits actually cost?

Usage credits are billed at standard API rates, which for Fable 5 means $10 per million input tokens and $50 per million output tokens. Two adjustments change what you actually pay. Cache reads bill at roughly 10% of the input rate, about $1 per million, and long Claude sessions are heavily cache-read. Cache writes bill at 1.25x the input rate, $12.50 per million, on the first pass of new context. And Fable 5's thinking is always on and billed as output, so the $50 side of the meter is where the money goes. Here is the arithmetic for three realistic days:

Light chat day. 30 substantive turns, each re-reading about 8K tokens of cached context, with roughly 30K tokens of fresh input and 20K tokens of output across the day. Cache reads: 240K tokens at $1 per million is $0.24. Fresh input at the cache-write rate: 30K at $12.50 per million is $0.38. Output: 20K at $50 per million is $1.00. Total: about $1.60. Call it $1.50 to $3 per day.

Heavy chat or coding day. A few hours of Claude Code style work, around 5M input tokens at 90% cache reads plus 150K output tokens including thinking. Reads: 4.5M at $1 per million is $4.50. Fresh input: 500K at $12.50 per million is $6.25. Output: 150K at $50 per million is $7.50. Total: about $18 per day.

All-day agent session. Multi-agent loops chewing through 10M input tokens at 90% cache reads with 1M output tokens. Reads: $9. Fresh input: $12.50. Output: $50. Total: about $70 per day, and it scales linearly from there. For calibration, the code-tester agent behind the Cheat Sheet processes about 40M input and 8M output tokens a week; at Fable 5 rates that is roughly $800 a week, which is exactly why it runs on Opus 4.8 instead. Metered Fable 5 is a scalpel, not a daily driver.

Do credits expire, and how does auto-reload work?

Anthropic's support article does not state an expiration period for purchased credits, so I will not invent one; treat your balance as spendable until Anthropic says otherwise, and do not stockpile more than a month or two of expected use. What the article does document: auto-reload automatically purchases funds when your balance drops below the threshold you set, it is optional, and it runs independently of the monthly spending cap. That last clause is the trap. The cap limits credit purchases per month, but if you set a generous cap and an aggressive reload threshold, the two together can authorize far more spend than you intended. Keep the cap tight and the reload off, and your worst month is a number you chose in advance.

Is Fable 5 worth paying for over Sonnet 5?

For most people, no, and that is not a consolation prize. Sonnet 5 is included in the plan you already pay for, it is genuinely strong at coding, writing, analysis, and agentic work, and it supports the same 1M token context window in chat on paid plans. The thing most people actually feel day to day, how much the model can hold at once, does not change when you decline to pay.

Pay for Fable 5 credits when both of these are true: the task is hard enough that model quality is the bottleneck, think long-horizon agent runs, gnarly multi-file refactors, reasoning problems Sonnet visibly fumbles, and the output is worth real money to you. A consultant whose deliverable bills at $2,000 should not blink at $15 of model spend. A student summarizing PDFs should not spend a cent. The honest middle path is the one I recommended before the deadline and still do: enable credits with a $10 or $20 cap and treat Fable 5 as an escalation button. Run everything on Sonnet 5, switch up only when Sonnet gets stuck. Metered pricing is bad at being a subscription and good at being a per-task decision, so make per-task decisions.

How do I make usage credits last longer?

Token discipline pays immediately once every wasted token has a price. Three moves cover most of the savings. First, keep a stable prompt prefix so caching does its 90% work; anything that changes at the top of your prompt invalidates the cache behind it, and the mechanics are in my prompt caching guide. Second, stop pasting whole documents when one section answers the question; input is cheap next to output, but a bloated context makes every turn cost more. Third, and biggest, write prompts that land on the first pass, because a re-run doubles the cheap input and the expensive output both. The full playbook, including where output tokens hide, is in my post on cutting AI token costs. If you want ready-made setups that already practice this, the 20 free agents at clskillshub.com/agents carry explicit token-budget instructions written for exactly this metered world.

What if I enabled credits and regret it?

Nothing dramatic is required. Turn off auto-reload so no new funds are purchased, let your remaining balance run out at its own pace, and switch your default model back to Sonnet 5 so the balance only drains when you deliberately pick Fable 5. Your subscription and included models are untouched by any of this; credits sit on top of the plan, they do not replace it. Anthropic's support article does not describe a refund path for unused credit balances, so plan to spend what you bought rather than counting on getting it back, which is one more argument for starting with a small amount.

One last thing before the meter starts mattering: make your prompts earn their tokens. The Cheat Sheet is 120 tested prompt patterns with the exact structure I use for each, built to get the right answer on the first pass and to cache well. When a sloppy Fable 5 re-run costs real money, first-time-right prompts pay for themselves on the first day. $15, everything included.