Claude Fable 5 Usage Credits: Plan Limits & Cost (2026)

Claude Fable 5 is the one Anthropic model that does not simply come with your subscription. Depending on which plan you are on, it is either included up to a hard ceiling, or it is metered separately as pay-as-you-go usage credits — and the rules changed more than once during 2026. This page is the current state as of late August 2026, checked against Anthropic's own support documentation and pricing pages rather than launch-day announcements.

What are Claude Fable 5 usage credits?

Usage credits are prepaid funds attached to a paid Claude plan that let you keep working after you hit your included usage limits, billed at standard Claude API rates. They are not a separate currency with its own exchange rate — a credit dollar buys exactly what an API dollar buys.

The mechanics, per Anthropic's usage credits documentation:

• Enable them at Settings → Usage. You add a payment method, set a monthly spend cap (or choose unlimited), then prepay by adding funds. Auto-reload will top up your balance when it falls below a threshold you set.
• Eligible plans are Pro, Max 5x, and Max 20x, plus Team and seat-based Enterprise where the organisation enables them. Mobile-app subscribers have to enable and buy credits on the web — it cannot be done inside the iOS or Android apps.
• Charges appear separately from your subscription on the bill, and there is a hard ceiling of $2,000 redeemable per day.
• Credits generally do not expire. The one exception is Japan, where credits purchased from 10 September 2026 expire six months after purchase, with a seven-day email warning first.
• Session limits still reset every five hours whether or not credits are enabled. Credits do not replace your plan allowance; they sit behind it.

Since May 2026 there is also a discounted variant. Usage bundles are credits bought in advance at a volume discount: a $50 bundle costs $45 (10% off), $250 costs $200 (20% off), and $1,000 costs $700 (30% off). Pro and Max subscribers can buy up to $2,000 of bundle value per month, Team owners up to $3,000. Bundle balance is one pool spanning Claude on web, desktop and mobile, Claude Code, Cowork, and third-party products signed in with your Claude account — and it only draws down once you are past your plan's included limits.

Which plans include Fable 5, and which bill it through credits?

This is the table most people are actually looking for. It comes from Anthropic's "Claude Fable 5 on your plan" article.

Two details in that table are widely misread.

The 50% figure is a ceiling, not a bonus. Anthropic is explicit that other models draw from the same weekly pool and that you can never exceed your weekly limit. So a Max 20x subscriber does not get "weekly limit plus 50% more of Fable 5" — they get the same weekly limit, of which at most half may be spent on Fable 5. Spend your week on Opus 5 and there is no separate Fable allowance waiting.

Pro subscribers are not locked out. Fable 5 works fine on Pro; it simply never touches the subscription allowance. Every Fable 5 token on Pro is a credit token. Eligible Pro and Team standard seats also received a one-time credit to soften the transition.

Fable 5 itself is available across Claude on the web, Claude Mobile, Claude Desktop, Claude Cowork, Claude Code, and Claude for Microsoft 365. Both your plan usage and your credit balance are visible in usage settings.

How much do Fable 5 usage credits actually cost?

Credits bill at standard API rates, which makes the cost precisely knowable. From Anthropic's pricing documentation:

Fable 5 is twice Opus 5 and five times Sonnet 5 per token. (Sonnet 5's $2/$10 is worth noting on its own: it launched as introductory pricing due to expire 31 August 2026, and Anthropic has since confirmed the planned increase to $3/$15 will not happen.)

What that means in practice, using arithmetic at published rates:

• A long research chat. Twenty turns where context grows to roughly 50K tokens, so cumulative input lands near 500K and output near 30K: about $5.00 input plus $1.50 output, so roughly $6.50 for the conversation — less if prompt caching is hitting, since cache reads bill at $1/MTok instead of $10.
• An hour of agentic coding. A Claude Code session that reads 2M input tokens with 80% served from cache and writes 100K output: 1.6M cache reads at $1 ($1.60), 400K fresh input at $10 ($4.00), 100K output at $50 ($5.00) — about $10.60.
• Occasional escalation. Reaching for Fable 5 a handful of times a day on genuinely hard problems, and staying on Opus 5 or Sonnet 5 otherwise, typically keeps the monthly credit spend in low single-digit dollars.

One factor that surprises people: Claude 4.7-and-later models use a newer tokenizer that produces roughly 30% more tokens for the same text than earlier models. Comparing a Fable 5 bill against an old Sonnet 4.6 bill therefore understates the gap — the price per token is higher and the same prompt counts as more tokens.

What changed on July 7, 2026, and what has changed since?

July 2026 is when Fable 5 stopped being free on subscriptions, and it is why so many people went looking for an explanation. The sequence was:

• Early July 2026. Fable 5 was available inside subscription limits under a launch promotion, capped at 50% of weekly usage. This is the period the "after July 7, 2026" searches refer to.
• 11:59:59 PM PT on 19 July 2026. The promotion formally ended. Anthropic's documentation gives this date, not July 7 — the earlier date circulated widely in coverage before the final cut-off was published.
• Since then. Fable 5 access split by plan tier along the lines in the table above: included with a 50% weekly ceiling on Max and premium seats, credits-only on Pro and standard seats, unavailable on Free.

A second, separate change landed a month earlier and is the one most likely to affect a developer's bill. As of 15 June 2026, Claude Agent SDK and claude -p usage no longer counts toward your plan's usage limits. Eligible Pro, Max, Team, and Enterprise subscribers can instead claim a separate monthly credit that covers the Agent SDK, claude -p, and third-party apps built on the Agent SDK. When that monthly credit runs out, further usage falls through to usage credits at API rates — and if credits are not enabled, requests simply stop until the monthly credit refreshes. Using the Agent SDK with a Claude Platform API key instead is unchanged: normal pay-as-you-go, and no monthly credit. We cover the fuller story in our write-up of Anthropic's June 2026 Claude Code billing change.

Do usage credits apply to Claude Code and the API?

This distinction confuses people constantly, so it is worth stating plainly. There are three billing worlds:

1. Interactive plan surfaces. Claude conversations on web, desktop and mobile, interactive Claude Code in your terminal or IDE, and Cowork all draw from one shared plan allowance. Anthropic is explicit that usage across every surface counts against the same limit — there is no separate Claude Code budget.
2. Usage credits. The overflow layer behind that allowance, priced at API rates, covering the same surfaces plus Research mode and project files.
3. The Claude API. A completely separate account and bill. API usage never touches your subscription limits, and your subscription never subsidises API calls. Fable 5 on the API is simply $10/$50 per million tokens, with a 50% discount via the Batch API ($5/$25).

The practical consequence: if a script, CI job, or internal tool is calling Claude, it is on API billing and none of the above credit rules apply. If you are typing into Claude Code, it is on plan billing, and credits are what keep you going past the limit.

Context windows follow the same surface split. In chat, Opus 5 and Sonnet 5 give 1M tokens on all paid plans while Opus 4.8, 4.7, 4.6 and Sonnet 4.6 give 500K. In Claude Code, Opus 5, 4.8, 4.7, Sonnet 5 and Fable 5 reach 1M on Pro, Max, Team and Enterprise — though Pro users need usage credits enabled to unlock that tier on Opus models.

What happens when you run out of usage credits?

Nothing breaks, and nothing is silently charged. The behaviour is layered:

• Hit your five-hour session limit with credits disabled: Claude pauses until the window resets. Session limits reset every five hours regardless of plan.
• Hit it with credits enabled and funds available: you are notified and can choose to continue on credits. It is an opt-in prompt, not an automatic switch.
• Hit your Fable 5 ceiling on Max or a premium seat: you can keep using Fable 5 on credits, or switch to another model and stay inside your plan allowance.
• Exhaust your credit balance or hit your monthly spend cap: usage stops until you add funds, raise the cap, or the window resets. Auto-reload exists precisely to avoid this mid-session.
• Hit your weekly limit: wait for the reset, upgrade the plan, or buy credits. Those are the only three options Anthropic lists.

Set a monthly spend cap before pointing any autonomous agent at Fable 5. An agentic loop that reads and rewrites a large codebase can consume millions of tokens in an afternoon, and at $50 per million output tokens the arithmetic moves quickly.

How can you make your Claude credits last longer?

Most credit spend is avoidable. In rough order of impact:

1. Pick the cheapest model that clears the bar. The single largest lever. Opus 5 is half Fable 5's price and Sonnet 5 is a fifth of it; if Opus 5 handles 90% of your work, make it the default and treat Fable 5 as an escalation path. Our comparison of Opus 5 versus Fable 5 covers where the gap is real and where it is not.
2. Lean on caching. Cache reads bill at 10% of input price ($1/MTok on Fable 5). Anthropic notes that project content is cached and does not count against limits when reused, and that similar frequently-used prompts get partial caching. Stable, reusable context is cheap; context you rebuild every turn is not.
3. Lower the effort level. Effort is one of the factors Anthropic lists as affecting consumption. Reasoning depth you do not need is tokens you paid for.
4. Keep conversations short and scoped. Every turn re-sends the whole history. A 50-turn thread is expensive by construction; starting fresh with a tight summary is not.
5. Trim what you attach. File length, artifacts, research mode, and web search all raise consumption. Research mode in particular is called out as consuming tokens more quickly.
6. Move automation to the API. Anything scripted, scheduled, or batchable belongs on API billing, where the Batch API halves the rate. Keep your plan allowance for interactive work.
7. Buy bundles if you spend consistently. If you reliably burn credits every month, the 10–30% bundle discount is free money.

For the terminal-specific version of this, see our guide to stretching Claude Code usage limits.

How does this compare to how OpenAI and Google meter their plans?

Anthropic's approach is unusual in being transparently priced. Because credits bill at published API rates, you can calculate a Fable 5 session's cost to the cent before you run it — the consumer plan and the developer platform share one price list.

Google takes the opposite approach on its AI plans. Chat access is described in relative multipliers rather than absolute quantities: AI Plus offers roughly 2× the free tier, AI Pro 4×, and AI Ultra 5× to 20× higher limits than Pro. Where Google does expose hard numbers, it uses a genuine credit pool — Flow gets 200 credits a month on Plus, 1,000 on Pro, and 10,000 on Ultra, with Flow Music at 3,000 / 10,000 / 30,000. Those credits are product-scoped allowances, not dollars, and there is no equivalent of topping up chat usage at API rates.

OpenAI publishes detailed API pricing — GPT-5.6 Sol at $4/$20 per million tokens on short context, Terra at $2/$12, Luna at $0.20/$1.20, and GPT-5.5 at $5/$30 — but its consumer plan metering is not documented alongside it, so there is no published mapping from a ChatGPT subscription to a token price the way there is with Claude credits.

The practical upshot for anyone budgeting: on a Claude plan you can forecast overflow spend arithmetically, which is exactly why setting a monthly cap is worth two minutes of your time.

FAQ

What are Claude Fable 5 usage credits?

Usage credits are prepaid pay-as-you-go funds on a paid Claude plan, billed at standard Claude API rates. On Pro and Team standard seats they are the only way to use Fable 5, because Fable 5 is excluded from those plans' usage limits. On Max and premium seats they cover Fable 5 usage beyond the included allowance.

How many credits do you get on each plan?

None by default — credits are purchased, not granted. What varies by plan is how much Fable 5 you get before credits kick in: nothing on Pro and Team standard seats, and up to 50% of your weekly usage limit on Max 5x, Max 20x, and premium Team or Enterprise seats. Fable 5 is unavailable on Free.

What changed on July 7, 2026?

Early July 2026 was the tail end of the launch promotion that made Fable 5 available inside subscription limits. Anthropic's documentation puts the formal end at 11:59:59 PM PT on 19 July 2026, after which Fable 5 became credits-only on Pro and standard seats and stayed included, capped at 50% of weekly limits, on Max and premium seats.

What happens when you run out of usage credits?

Usage stops until you add funds, raise your monthly spend cap, or your window resets — nothing is charged silently. Your subscription itself keeps working with the other models, and five-hour session limits continue to reset normally. Auto-reload can top the balance up automatically if you would rather not be interrupted.

Do usage credits apply to the API?

No. The Claude API is billed separately and never touches your subscription or credit balance. Fable 5 on the API costs $10 per million input tokens and $50 per million output, halved to $5/$25 through the Batch API. Credits apply only to plan surfaces: Claude on web, desktop and mobile, Claude Code, Cowork, and Claude for Microsoft 365.

How can I make my Claude credits last longer?

Default to a cheaper model and escalate to Fable 5 deliberately, since Opus 5 is half the price and Sonnet 5 a fifth. Reuse cached project context, which bills at 10% of input rates. Lower the effort level, keep threads short, trim attachments, and move scripted or batchable work onto the API where batching halves the rate.

The decision rule

If you are on Pro and reaching for Fable 5 daily, do the arithmetic before enabling credits: at $10/$50 per million tokens, regular agentic use will comfortably exceed the $100 gap between Pro and Max 5x, and Max gives you Fable 5 inside the plan up to half your weekly limit. If you reach for Fable 5 a few times a week on genuinely hard problems, Pro plus a small credit balance and a monthly spend cap is the cheaper shape. Either way, set the cap first — the failure mode here is not being denied access, it is an unattended agent quietly spending real money at frontier-model rates.

For the model itself, see our Claude Fable 5 launch guide, and for the wider landscape of terminal agents and how they bill, the AI coding agents complete guide.