As of July 21, 2026, Claude Fable 5 does not have one universal subscription cost. On Max plans and Premium seats on Team or seat-based Enterprise, Fable can use up to 50% of the plan or seat's existing weekly limit at no extra charge. On Pro and Team Standard, Fable uses pay-as-you-go usage credits from the first Fable request. Standard seats on legacy seat-based Enterprise also need organization-enabled credits.
When Fable is metered, the base rate is $10 per million input tokens plus $50 per million output tokens. A task with 50,000 input tokens and 10,000 output tokens therefore consumes about $1.00 of usage-credit balance before cache, region, feature, tax, or marketplace adjustments.
| Your route | What Fable 5 uses first | What happens next |
|---|---|---|
| Max 5x or Max 20x | The normal weekly plan pool, up to a 50% Fable share | Switch models or continue with paid usage credits after that Fable share is reached |
| Team Premium or eligible Enterprise Premium seat | The seat's normal weekly pool, up to a 50% Fable share | Organization policy and credits control paid continuation |
| Pro or Team Standard | Usage credits from the first Fable task | Balance falls at standard API rates |
| Legacy seat-based Enterprise Standard | Organization-enabled usage credits from the start | Access stops if the organization does not enable credits |
| Usage-based Enterprise or Claude API | Metered billing at standard API rates | Project or organization billing owns the spend |
The phrase “up to 50%” does not add 50% to a weekly limit. Anthropic's current Fable plan guide says Fable and other models draw from the same weekly pool. The percentage is a ceiling on how much of that pool can go to Fable.
Calculate Fable 5 cost from tokens
Use the standard uncached formula:
textcost = input_tokens / 1,000,000 × $10 + output_tokens / 1,000,000 × $50
Three reproducible examples make the scale clearer:
| Workload meter | Input cost | Output cost | Base total |
|---|---|---|---|
| 10,000 input + 2,000 output | $0.10 | $0.10 | $0.20 |
| 50,000 input + 10,000 output | $0.50 | $0.50 | $1.00 |
| 200,000 input + 40,000 output | $2.00 | $2.00 | $4.00 |
These are meter examples, not promises about how many messages a credit balance buys. In Claude or Claude Code, the next turn may reprocess conversation history, project instructions, retrieved files, tool results, or other context. A retry can repeat much of that input. A long answer is especially expensive because the output rate is five times the input rate.
For workloads explicitly using US-only inference, Anthropic currently lists a 1.1x multiplier on both input and output. The $1.00 example becomes $1.10 and the $4.00 example becomes $4.40. Do not apply that multiplier merely because the user is in the United States; it belongs to the US-only inference contract.
The official Fable model page and Claude pricing table list five-minute prompt-cache writes at $12.50 per million tokens, one-hour writes at $20 per million, and cache reads at $1 per million. Caching can materially lower repeated-input cost, but only when the route actually reuses eligible cached content. It does not discount generated output.
Fable 5 also uses Anthropic's newer tokenizer. The official token-counting guide says the same text produces roughly 30% more tokens than on models before Claude Opus 4.7, with the exact change depending on content and workload shape. Treat that as a migration warning, not a universal multiplier: count the real request with model: "claude-fable-5" or budget from observed usage records.
What a Claude “usage credit” means
A usage credit is account value, not a fixed message or token coupon. Anthropic's paid-plan usage-credit guide says the balance is prepaid and consumed at standard API rates. It appears separately from the subscription charge and can be controlled with a monthly cap, alerts, and optional auto-reload.
That creates three meters that should not be mixed:
- Included plan usage belongs to the Claude subscription or seat and resets under the plan's session and weekly rules.
- Paid-plan usage credits continue eligible Claude, Claude Code, and other product use when the plan says a task is metered.
- Claude Console credits fund API and Workbench activity for a Console organization or project.
The same dollar price row can apply to the second and third meters, but they still have different owners and dashboards. A Pro subscriber using Fable in Claude Code should inspect Settings > Usage. An application, CI job, eval runner, or service should inspect Claude Console billing. If Claude Code's active route is unclear, use the separate API-key versus subscription billing check before buying more balance.
Who gets the one-time $100 Fable credit
Anthropic currently offers a time-limited transition promotion for accounts that qualified before Fable moved to the July 20 plan rules. The official one-time Fable credit article lists:
- Pro: $100 of one-time usage credits for eligible monthly or annual subscribers.
- Team Standard: $100 per purchased qualifying Standard seat, pooled at the organization level, up to $2,500.
- Max, Team Premium, and Enterprise: not eligible for this particular promotion because Max and Premium retain included Fable access, while Enterprise has separate terms.
Eligibility depends on holding the qualifying subscription or purchased seat at July 19, 2026, 11:59:59 PM PT. Free trials do not qualify. The current claim deadline is August 2, 2026, 11:59 PM PT, and the promotional balance expires September 17, 2026, 11:59 PM PT regardless of when it was claimed.
The $100 is not Fable-only money. It is standard usage-credit balance and can be spent on other eligible models. Claiming it enables usage credits and may ask for a payment method, but Anthropic says claiming by itself does not charge the account. Later top-ups or auto-reload are separate actions.
This offer is temporary. Check the live help page and the amount shown in your own Settings > Usage before relying on it. After the claim window closes, an old screenshot or search snippet is not evidence that a new subscriber can still receive it.
Bundles discount the balance purchase, not the token meter
Eligible Pro, Max, and Team accounts can also buy discounted usage bundles. Anthropic's bundle guide currently lists these face values:
| Balance added | Checkout price | Cash discount | Important boundary |
|---|---|---|---|
| $50 | $45 | 10% | Fable still draws against the $50 face-value balance at the standard meter |
| $250 | $200 | 20% | The bundle does not enlarge the weekly subscription limit |
| $1,000 | $700 | 30% | Availability, account controls, and monthly bundle caps still apply |
If the earlier $1.00 task is paid entirely from a $250 bundle bought for $200, its underlying balance drawdown remains $1.00, while its effective cash acquisition cost is about $0.80. This distinction prevents a common accounting error: claiming the model rate changed when only the prepaid balance was discounted.
Pro and Max accounts currently have a $2,000 monthly bundle-value cap; Team has $3,000. Usage beyond the discounted-bundle cap can still run at standard usage-credit rates if the account allows it. Set a monthly limit before enabling auto-reload, especially on Pro or Team Standard where Fable starts on the paid meter.
Why a real Claude Code session can drain faster
The arithmetic is simple, but the workload is not. Four factors usually explain why observed spend exceeds prompt-only math.
Conversation growth. A long thread can carry earlier instructions, code, file excerpts, and tool results into later turns. The visible new prompt may be short while billed input is not.
Agent loops and retries. Claude Code may inspect files, run commands, read errors, revise a plan, and generate a patch. Those steps create additional input and output. A failed first attempt still used model work even if only the final patch was valuable.
Output length. At $50 per million output tokens, verbose explanations and large generated files can dominate the bill. Ask for the artifact and verification that the task needs, not an unlimited narrative.
Cache behavior. Repeated stable context can be cheap when it becomes an eligible cache read. Rewritten prompts, changing project context, or a route that does not reuse the cache will not receive the same benefit.
Do not turn those variables into a universal “messages per $100” number. A short isolated analysis and a multi-hour repository migration can both count as one user request while consuming radically different token volumes.
Decide between included Fable, credits, and a cheaper model
Use Fable where a difficult task can justify the premium: long-running engineering, complex document work, or a high-value problem where fewer failed attempts may offset the higher token rate. Use a cheaper model for routine editing, classification, simple summaries, or repeated low-risk steps unless a controlled test shows Fable produces more accepted work.
The plan decision is route-specific:
- Already on Max or Premium: use the included Fable share first, but monitor Settings > Usage because the 50% ceiling and the whole weekly pool still apply.
- On Pro or Team Standard: set a small credit cap, run one representative task, and record the balance change before enabling a large bundle or auto-reload.
- Building an app or automation: use a Console/API project with an explicit budget. A personal subscription is not the owner of production API work.
- Hitting credits every week: compare the actual accepted-work cost with Max/Premium and cheaper-model routes. The broader Claude subscription versus API guide covers that workload-owner decision.
A practical stop rule is: do not buy the next block of credits until the last block produced accepted work that a cheaper route did not provide at lower total cost. Include retries and review time, not only token spend. If Fable is impressive but the task would have passed with Sonnet or Opus, route the ordinary steps down and reserve Fable for the difficult branch.
Verify before the next paid run
Before sending a long Fable task, check:
- your exact plan and seat type in Settings;
- whether Fable is using the included weekly pool or usage credits from task one;
- remaining Fable share and overall weekly usage;
- credit balance, monthly cap, and auto-reload state;
- whether a temporary promotion or bundle actually appears in the account;
- the current Fable input, output, cache, and region rates;
- for Claude Code, version 2.1.170 or later and the active billing route;
- for API work, the project budget and Console usage rather than the consumer plan page.
This source-of-truth sequence matters more than any viral per-message estimate. Fable 5 can be included, metered from the first task, or billed as API usage depending on the route. Identify that route first, apply the token formula second, and only then decide whether the result is worth another credit.