Claude Fable 5 API Pricing: Token Costs, Usage Credits, and Cost Examples
Anthropic's direct 5-minute prompt-cache write costs $12.50 per million tokens, a 1-hour cache write costs $20 per million, and cache hits or refreshes cost $1 per million. Batch processing reduces direct base input and output rates by 50%. Fable's full 1M-token context window uses standard per-token rates rather than a separate long-context multiplier.
Claude Fable 5 API Pricing at a Glance
| Billable item | Official price | What it covers |
|---|---|---|
| Base input | $10 / 1M tokens | Uncached prompt, system, message, tool-schema, and context tokens |
| 5-minute cache write | $12.50 / 1M tokens | Tokens written to the short prompt cache |
| 1-hour cache write | $20 / 1M tokens | Tokens written to the extended prompt cache |
| Cache hit or refresh | $1 / 1M tokens | Reused prompt-cache tokens |
| Output | $50 / 1M tokens | Model-generated thinking and response tokens billed as output |
| Batch input | $5 / 1M tokens | Asynchronous Batch API base input |
| Batch output | $25 / 1M tokens | Asynchronous Batch API output |
The output rate is five times the base input rate. That means long responses, retries, and agent verification loops can matter more than the original prompt size.
Subscription Access vs API Billing
Fable 5 has three cost surfaces that should not be mixed together.
| Access path | Billing model | What to verify |
|---|---|---|
| Claude paid-plan promotion | Temporarily included in eligible plan limits | Promotion ends July 12, 2026 at 11:59:59 PM PT |
| Claude usage credits | Prepaid consumption billed separately | Anthropic says standard API rates apply |
| Direct or gateway API | Per-token API billing | Provider, cloud, gateway, cache, and route-specific terms |
During the promotion, eligible users can spend up to 50% of weekly plan limits on Fable 5. After the deadline, Fable no longer draws from the included weekly allowance. Enabling usage credits does not convert the subscription into an unlimited Fable plan; it adds consumption charges on top of the subscription.
claude-fable-5 through an API, the promotion deadline does not change the basic token-billing model.Fable 5 Cost Formula
For a standard uncached request:
cost = (input_tokens / 1,000,000 × $10)
+ (output_tokens / 1,000,000 × $50)For a request with prompt caching:
cost = (fresh_input_tokens / 1,000,000 × $10)
+ (cache_write_tokens / 1,000,000 × cache_write_rate)
+ (cache_hit_tokens / 1,000,000 × $1)
+ (output_tokens / 1,000,000 × $50)$12.50 for 5-minute cache writes or $20 for 1-hour cache writes. A request normally has either cache-write or cache-hit tokens for a specific cached block, not both at the same time for the same block.For server-side tools, add the documented tool charge. Tool definitions and tool results also consume normal input tokens. Client-side tools do not have a separate Anthropic tool fee, but their schemas and results still increase token usage.
Worked Fable 5 Cost Examples
These examples use official base rates only. They exclude taxes, enterprise agreements, provider-specific fees, web-search calls, retries beyond those stated, and EvoLink route discounts.
| Workload | Token assumption | Calculation | Cost |
|---|---|---|---|
| Focused code review | 50K input + 5K output | 0.05 × $10 + 0.005 × $50 | $0.75 |
| Repository agent turn | 200K input + 20K output | 0.2 × $10 + 0.02 × $50 | $3.00 |
| Long-context analysis | 1M input + 50K output | 1 × $10 + 0.05 × $50 | $12.50 |
| Ten repository turns | 2M total input + 200K total output | 2 × $10 + 0.2 × $50 | $30.00 |
The ten-turn example is deliberately simple. Real coding agents can resend history, tool schemas, file contents, and tool results on each turn. Measure the API's reported usage instead of multiplying the first prompt size by the number of turns.
Prompt Caching: First Request vs Repeated Requests
Prompt caching changes the cost shape when a large, stable prefix is reused.
Assume a request contains:
- 500K tokens of stable repository context
- 50K tokens of fresh instructions and tool results
- 20K output tokens
- a 5-minute cache
The first request creates the cache:
500K cache write × $12.50 / 1M = $6.25
50K fresh input × $10 / 1M = $0.50
20K output × $50 / 1M = $1.00
first request total = $7.75A repeated request that hits the cache costs:
500K cache hit × $1 / 1M = $0.50
50K fresh input × $10 / 1M = $0.50
20K output × $50 / 1M = $1.00
repeat request total = $2.00In this example, the repeated request is $5.75 cheaper than rebuilding the same 5-minute cache. That does not mean caching always saves money. The cached prefix must remain stable long enough and be reused often enough to recover the higher write cost.
For a 5-minute cache, the cache-write premium over base input is $2.50 per million tokens, while each hit saves $9 per million compared with fresh input. One valid hit is enough to recover the write premium for the reused tokens. A 1-hour cache has a higher write cost and should be justified by a longer reuse window.
Long Context Does Not Add a Fable 5 Multiplier
Anthropic states that Fable 5 includes the full 1M-token context window at standard pricing. A 900K-token request uses the same per-token rates as a 9K-token request.
That does not make long context inexpensive. At the official base rate:
- 100K uncached input costs $1
- 500K uncached input costs $5
- 1M uncached input costs $10
The cost risk comes from repeatedly resending large context across agent turns. Retrieval, compaction, and prompt caching usually matter more than the single-request context limit.
Batch API Pricing
Anthropic's Batch API reduces base input and output prices by 50%:
- Fable 5 batch input: $5 / 1M tokens
- Fable 5 batch output: $25 / 1M tokens
Batch is appropriate for asynchronous work that does not need an immediate response, such as offline evaluation, large-scale classification, scheduled analysis, or regression testing. It is not a replacement for interactive Claude Code or latency-sensitive agents.
Prompt caching modifiers can apply with Batch API pricing according to Anthropic's current pricing rules. Verify the exact combination before calculating a large job.
US-Only Inference Pricing
- US-only base input: $11 / 1M tokens
- US-only output: $55 / 1M tokens
This arithmetic applies the documented 1.1 multiplier to the official base rates. Confirm platform availability, caching treatment, and any other modifiers before using it in a contract or production budget.
Fable 5 vs Opus 4.8 Cost
Anthropic lists Opus 4.8 at $5 per million base input tokens and $25 per million output tokens. Fable 5 is exactly twice those standard token rates.
| Example | Opus 4.8 | Fable 5 | Fable premium |
|---|---|---|---|
| 50K input + 5K output | $0.375 | $0.75 | $0.375 |
| 200K input + 20K output | $1.50 | $3.00 | $1.50 |
| 1M input + 50K output | $6.25 | $12.50 | $6.25 |
Fable becomes cost-effective only when its quality reduces a different cost: failed agent runs, repeated prompts, senior review, production defects, or delayed completion. Compare the premium with the value of those avoided failures.
Production Budget Guardrails
Use these controls before enabling Fable for production traffic:
- keep Fable behind an explicit task-difficulty rule
- cap input and output tokens per request
- cap retries and agent steps
- cache stable context when reuse is measurable
- log fresh input, cache write, cache hit, and output separately
- record actual model and refusal or fallback behavior
- set a monthly usage-credit or gateway spend cap
- compare cost per accepted task with Opus 4.8 and Sonnet 5
- alert on unusually large prompts, outputs, or repeated failures
Do not estimate a production budget from a single successful demo. Use several weeks of representative traces and include failed runs.
EvoLink's Current Fable 5 Discount
- $9 / 1M input tokens, versus Anthropic's $10 list rate
- $45 / 1M output tokens, versus Anthropic's $50 list rate
- $11.25 / 1M cache-write tokens, versus Anthropic's $12.50 5-minute cache-write rate
- $0.90 / 1M cache-hit tokens, versus Anthropic's $1 cache-hit rate
$2.70 at the current EvoLink route rates instead of $3.00 at direct list pricing, before any other route, account, or tool charges.The arithmetic is:
200K input × $9 / 1M = $1.80
20K output × $45 / 1M = $0.90
EvoLink route total = $2.70
Direct list total = $3.00
Savings = $0.30 (10%)EvoLink pricing can change independently from Anthropic's list price. This article records the verified snapshot, while the existing live price module remains the source of truth for the current route.
View Current Fable 5 Pricing on EvoLinkThe model page shows the current EvoLink input, output, cache-write, and cache-hit rates. Verify the displayed rate and account terms before moving production traffic.
Sources
- Anthropic: Claude Platform pricing
- Anthropic: Claude Fable 5 product page
- Anthropic: Fable 5 and Mythos 5 API behavior
- Anthropic: Fable 5 promotional access
- Anthropic: Manage usage credits for paid plans
- EvoLink: Claude Fable 5 API and current route pricing
FAQ
How much does Claude Fable 5 API cost?
Anthropic's official list price is $10 per million base input tokens and $50 per million output tokens. Prompt caching, Batch API, server-side tools, and data-residency options can change the final charge.
How much do Fable 5 usage credits cost?
Anthropic says usage credits are billed at standard API rates and charged separately from the paid Claude subscription. Users prepay funds and can set a monthly limit.
Is Fable 5 included in Claude Max after July 12, 2026?
No. Anthropic says the promotion ends July 12, 2026 at 11:59:59 PM PT. Continued Fable usage requires separately billed usage credits or API access.
What is the Fable 5 cached-input price?
Cache hits and refreshes cost $1 per million tokens. A 5-minute cache write costs $12.50 per million, while a 1-hour cache write costs $20 per million.
Does Fable 5 charge more above 200K context?
No separate long-context multiplier is documented for Fable 5. Anthropic says the full 1M-token context window uses standard per-token pricing.
Is Fable 5 exactly twice the price of Opus 4.8?
Yes for official standard base input and output rates. Fable is $10/$50 per million input/output tokens, while Opus 4.8 is $5/$25.
How can I reduce Fable 5 API cost?
Route only hard tasks to Fable, cap output and retries, cache stable context, use Batch for asynchronous jobs, compact long histories, and measure cost per accepted task.
Does EvoLink use the same price as direct Anthropic?
Not necessarily. EvoLink route pricing is shown in the live pricing module on the Fable 5 model page. Use the official Anthropic rate as a comparison baseline and the live EvoLink rate for current route decisions.
