Claude Fable 5 Pricing Explained: API Rates, Caching, Plans

Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens on the Claude API - exactly twice Opus 4.8, and less than half what Claude Mythos Preview cost its ~50 partners. Whether that is expensive depends entirely on how you use it. This guide lays out every published rate, the claude.ai subscription rules, real-world cost data from the first day of usage, and the levers that bring bills down.

The headline rates

Model	Input / MTok	Output / MTok	Relative to Fable 5
Claude Fable 5	$10.00	$50.00	-
Claude Opus 4.8	$5.00	$25.00	1/2 the price
Claude Sonnet 4.6	$3.00	$15.00	~1/3 the price
Claude Haiku 4.5	$1.00	$5.00	1/10 the price

Two details worth noting. First, the 1M-token context window carries no long-context premium - rare at the frontier. Second, requests pinned to US-only inference carry a 1.1x multiplier on these rates. Full rate cards are on platform.claude.com.

Batch and prompt caching

The Batch API halves everything: $5 per million input tokens and $25 per million output tokens for asynchronous workloads that can tolerate delayed results. That puts batched Fable 5 at exactly Opus 4.8's real-time price.

Prompt caching is where heavy users save the most:

Caching tier	Price / MTok	vs standard input
5-minute cache write	$12.50	1.25x
1-hour cache write	$20.00	2x
Cache hit (read)	$1.00	90% discount

The minimum cacheable prefix is 512 tokens (1,024 on Amazon Bedrock). For agents that re-read a large system prompt or codebase context on every turn, paying $1 instead of $10 per million on repeated tokens changes the economics of the whole workload.

Fable 5 in claude.ai plans

Inside Claude subscriptions, Fable 5 counts roughly 2x the usage of Opus toward your plan limits - the price ratio carried over into the metering. The launch terms, per Anthropic's announcement:

June 9-22, 2026: Fable 5 is free in Pro, Max, Team, and Enterprise plans.
After June 22: Fable 5 requires usage credits billed at API rates.
Anthropic says it "aims to restore Fable 5 as a standard part of subscription plans" once inference capacity allows.

Try it now: the free window is the cheapest Fable 5 will be for the foreseeable future. If you are on any paid plan, the next two weeks are the time to test it on your real workloads.

The effective-cost argument

Per-token price is the wrong number to fixate on if the model finishes work in fewer tokens. A Canva engineer made this case on Hacker News: in their testing, Fable 5 produced better results with roughly half the tokens Opus 4.8 used - fewer retries, less wandering, tighter output. Half the tokens at twice the price is cost parity, with a better result. The same logic shows up in Amazon's launch quote ("more capable engineering in fewer turns"). Our Fable vs Opus comparison works through the math benchmark by benchmark.

What a real day costs

Simon Willison published his numbers from a full day of intensive Fable 5 use across experiments and real work: $110.42. Heavy, but in line with a day of frontier-model agentic coding - and his per-request breakdown shows how much the effort setting moves the bill:

Effort setting	Output tokens (same prompt)	Cost
low	1,929	$0.10
max	14,430	$0.72

A 7x spread in output tokens - and cost - from one parameter, on one prompt. Since Fable 5's adaptive thinking is always on, effort is the primary dial you control; see our effort parameter guide for when each level pays off.

Four levers for controlling costs

Tune effort per task. Default to medium or high; reserve xhigh and max for problems that justify a 5-7x output multiple, and drop to low for routine transformations.
Batch what can wait. The Batch API's 50% discount makes overnight evaluation runs, backfills, and bulk processing cost the same as real-time Opus 4.8.
Cache aggressively. Structure prompts so stable context (system prompt, docs, code) forms a prefix of at least 512 tokens, then pay $1/MTok on every hit instead of $10.
Set task budgets. For long-horizon agent sessions, cap spend per task and let the model work within it rather than discovering the bill afterward.

The bottom line

At list price, Fable 5 is a premium model: 2x Opus 4.8 on every token. In practice, the gap narrows or disappears for workloads where it solves problems in fewer tokens and fewer attempts - and batch plus caching can pull effective rates below what you were paying Opus. Measure your own token consumption during the free window before deciding; for routine, cost-sensitive work, Opus 4.8 and Sonnet 4.6 remain the sensible defaults.