Plan rules move — verify before you rely on them
Usage-credit mechanics and the Fable 5 inclusion rules changed in July 2026 — Fable 5 is now included on Max and Team Premium for up to 50% of weekly usage limits. The Max plan page tracks the current inclusion and limit rules.
See the Max plan pageClaude Max & Pro · billing guide
Claude Fable 5 usage credits, explained
The same model, two very different bills. Here is what usage credits are, why your Max allowance drains so fast, and the settings that stop a surprise charge.
TL;DR — 30 seconds
Usage credits are consumption-based billing that kicks in on paid Claude plans (Pro, Max 5x, Max 20x) after your included allowance is used up. They bill at standard API rates — currently $10 per million input tokens and $50 per million output tokens for Fable 5 — and are charged separately from your subscription, so you are buying tokens, not another plan. Credits are entirely separate from a direct API account: code you run through the API is billed per token on that account, never from your Claude plan balance. On Max, Fable 5 is included for up to 50% of your weekly pool (as of July 20, 2026) — beyond that, or on Pro/Team Standard, Fable 5 runs on credits. Everything is capped where you would expect: Settings → Usage gives you a monthly spending cap, auto-reload and usage alerts. Set the cap before you start a long agent run, not after.
What usage credits are (and are not)
One model, two billing lanes that never cross.
On a plan: credits are an overflow lane
Pro, Max 5x and Max 20x all come with an included weekly allowance. Once it is exhausted, usage credits let you keep going at standard API rates instead of being blocked. Your session limits still reset every five hours, and credits are charged separately from your subscription.
The API is a different account
Direct API usage is metered per token on your API account — input, output and cache reads at published rates. A Claude plan balance never pays for API calls, and an API key never spends your plan's credits. Check which surface you are on before you estimate cost.
What credits are for
Anthropic describes credits as the seamless way to continue after your included limit — the price of staying in the flow on a model that is token-hungry. On Pro and Team Standard, Fable 5 generally runs on credits from the first request.
Why the allowance drains so fast
Fable 5's weekly share is a cap, and several mechanics push usage toward it at once.
Tokens are the only currency
Everything you send and receive is billed: input, output, prompt-cache reads, and — on Fable 5 — the reasoning tokens the model thinks before it answers. Longer inputs and longer answers both count.
Claude + Claude Code count together
Your combined usage across conversations and Claude Code terminal sessions counts toward the same limits, so a day of coding burns the same pool as a day of chat.
Research mode and big contexts amplify everything
Anthropic notes that research sessions consume tokens more quickly — multiple searches and long analyses. Attached project files and documents count toward context too: every token processed, including project content, is billed.
Effort multiplies the output side
Fable 5's effort setting controls how much reasoning the model spends. Higher levels push output tokens well above the high baseline (estimated 1.45x at xhigh and 1.9x at max), and output is the expensive direction at $50 per million tokens.
Agent runs loop the meter
Autonomous Claude Code sessions re-read the same context on every step. A single long task can land dozens of requests with a large context window, and each one bills input plus output — this is the fastest way to reach a weekly cap.
How much a typical task costs
Credit consumption follows API token math, so these are estimates at Fable 5's standard rates ($10/$50 per million tokens). Effort changes the output column; caching, context size and agent loops change the input side. Your numbers will differ — this shows the scale.
| Task | High effort | Max effort |
|---|---|---|
| Analyze a 200K-token document | ≈$1.16 | ≈$1.34 |
| Analyze a 1M-token context | ≈$5.41 | ≈$5.95 |
| One agent run, 20 steps | ≈$5.85 | ≈$7.65 |
Estimates from the same estimator as the cost calculator, for a single Fable 5 request. Long-doc: 200K input tokens (70% from cache), 4K output at high. 1M context: 1M input (70% cached), 12K output. Agent: 20 steps × 40K input (70% cached) + 2K output each, high = 1× and max = 1.9× output multipliers. One step of a 1M-context run bills more input than most standalone requests.
One long agent day can clear a meaningful share of a Max weekly pool on its own. A daily redemption limit of $2,000 applies to usage credits.
Monitoring and spending caps
All from Settings → Usage in the Claude app — you do not need a separate tool.
Usage dashboard
Real-time consumption, month-to-date credit spend and usage history, with your included plan allowance and any usage-credit consumption shown separately. This is the first place to look when something feels fast.
Monthly spending cap
Set a maximum you are willing to spend on usage credits each month. It is the one setting that makes a runaway agent run cost something instead of anything.
Auto-reload
Automatically tops up when your balance falls below a threshold you choose — convenient, but it can keep spending past the point you meant to stop. A cap is still the safer default.
Usage alerts
Notifications when you approach your spend limits. Cheap to turn on and worth it: the first alert is usually when people discover how fast research mode eats tokens.
Know what you can change
You can cap or disable usage credits entirely — after that, you only have your plan's included usage. The 50% Fable share on Max is set by the weekly limit, not by a credit setting. Anthropic also reserves the right to limit usage through weekly and monthly caps at its discretion.
Max 5x vs Max 20x: what the allowance covers
Two worked examples with real workload numbers, at Fable 5's API rates. Weekly plan limits are expressed in relative terms (5x and 20x vs Pro) and are not published in hours or tokens, so treat the absolute figures as estimates.
Max 5x ($100/month)
5× the usage of Pro per session. At Fable 5 rates, a single 200K-document analysis is about $1.15–1.35, a 1M-context analysis about $5.50–6, and a continuous hour of 200K-context analysis roughly $70 in tokens. A 20-step agent run that re-reads a 1M context each step lands around $100. Anthropic publishes the multiplier but not the pool size, so a heavy day can press against the weekly pool.
Max 20x ($200/month)
20× Pro — 4× the 5x pool at 2× the price, which makes it the cheaper per-unit tier for heavy use. A week of agent-heavy work (say 5–10 runs of 1M-context tasks, roughly $500–1,000 of Fable 5 tokens) fits far more comfortably here. When the pool is gone, usage credits pick up at the same rates.
Weekly limits are published only as relative multipliers (5×, 20× vs Pro) — Anthropic publishes no hour or token figures. How far a pool stretches depends on your mix of context size, caching and effort; treat every absolute number above as an order-of-magnitude estimate. The $2,000 daily redemption cap and your monthly spending cap apply to usage credits.
Estimate your own workload
The cost calculator runs Fable 5's token math with your inputs — context size, caching, effort level and fallback share — before you commit credits to a run.