How Much Does Claude Fable 5 Cost?

Published on June 9, 2026 | Written by Alec Pow
This article was researched using 9 sources. See our methodology and corrections policy.

Claude Fable 5 is Anthropic’s widely released Mythos-class model for long-running AI work. The public cost starts with the API meter: as of June 2026, Anthropic lists Fable 5 at the Fable product sheet for $10 per million input tokens and $50 per million output tokens, with higher or lower totals depending on app plan, cloud route, cache use, tools, and output length.

The bill has two paths. Claude app users pay for access limits through Free, Pro, Max, Team, or Enterprise. Developers and usage-based Enterprise customers pay by tokens, cache events, tool calls, managed-agent runtime, and cloud-marketplace metering.

Fable 5 is priced by workload unit, not just by seat. A short chat, a long codebase migration, a research agent, and a cached document workflow can all use the same model, but the input-output split and tool choices can make their monthly totals very different.

TL;DR: Fable 5 starts at $10 per million input tokens and $50 per million output tokens, but real monthly spend depends on whether you use it through Claude plans, the API, or cloud marketplaces.

How Much Does Claude Fable 5 Cost?

Jump to sections

• API rate: $10 (equivalent to 20 minutes at $30 per hour, or about $4 in 1990 dollars) per MTok input and $50 per MTok output.
• Cache rate: Fable 5 cache hits are $1 per MTok on Anthropic’s model pricing table.
• Batch rate: Fable 5 batch processing is $5 per MTok input and $25 per MTok output.
• App access: Claude Pro is $17 monthly with annual billing, $20 monthly, and Max starts at $100 per month on the current plan grid.

Who this cost makes sense for

Fable 5 makes sense when a buyer is paying for a hard reasoning model rather than a casual chat plan. The right use case is long coding work, analysis across many files, document-heavy finance or legal review, autonomous agent sessions, or research tasks where a weaker model creates too many retries.

The weak case is volume work with short, repeated outputs. Classification, basic extraction, support triage, and routine summaries can often run on Sonnet or Haiku at lower token rates. A team should also cap output length, cache repeated prompts, and track tool use before moving Fable 5 into default workflows.

Makes sense if

• The task needs Fable 5’s long-horizon reasoning.
• Better first-pass output cuts human review time.
• The workflow can use cache reads or batch jobs.
• Admins can watch token use and tool charges.

Doesn’t make sense if

• The task is short triage or basic data cleanup.
• The project needs a fixed monthly chat bill only.
• Long outputs will run without length controls.
• Agent sessions can run with no spend cap.

What you’re actually buying

Claude Fable 5 is a high-capability Claude model used through the Claude app, Claude API, and supported cloud platforms. It is built for demanding reasoning, coding, analysis, visual document work, and agent workflows where a cheaper model may need extra prompting or closer review.

It is not the same purchase as Claude Pro or Team by itself. App plans control access and usage limits, while API customers pay by token usage. It also differs from Claude Mythos 5, which Anthropic describes as restricted to approved Project Glasswing users rather than broad self-serve access.

How Fable 5 billing works in the API

The API charge has two core lines: input tokens and output tokens. Anthropic says Fable 5 and Mythos 5 are offered at $10 (about $4 in 1990 dollars) per million input tokens and $50 per million output tokens in its June 2026 launch note. Output costs five times input at the base rate, so verbose answers and generated code files can drive the bill faster than long prompts alone.

Prompt caching changes the math. Anthropic lists Fable 5 cache writes at $12.50 per MTok for five minutes, $20 per MTok for one hour, and cache hits at $1 per MTok on its prompt caching details. That rewards repeated system prompts, shared documents, and recurring project files.

Billing line	Public rate	Budget meaning
Base input	$10 per MTok	Prompt, files, tool schemas, and retrieved text
Base output	$50 per MTok	Answers, code, reports, and agent notes
Cache hit	$1 per MTok	Repeated context after a cache write
Batch input and output	$5 and $25 per MTok	Asynchronous work that can wait

The tiers, renewals, and platform route

Claude app buyers should separate subscription access from metered model use. Pro and Max are individual access plans. Team has Standard and Premium seats. Enterprise can mix a seat fee with API-rate usage, so the buyer is not only choosing a model, but also a billing route and control layer.

Team pricing can become the clearer company budget because seats are visible. Anthropic lists Team Standard at $20 (about $8 in 1990 dollars) per seat per month annually or $25 monthly, and Team Premium at $100 annually or $125 monthly. Teams comparing this with Claude business pricing should map each seat to a role, not give Premium to every user by default. For a broader AI chat benchmark, ChatGPT subscription prices can give a second monthly-plan reference.

Add-ons and hidden usage costs

The hidden-cost band for Fable 5 runs from small tool charges to agent sessions that stay active for hours. Web search adds $10 per 1,000 searches, code execution adds $0.05 per hour after 1,550 included monthly hours when used without web search or web fetch, and web fetch has no separate charge beyond tokens under Anthropic’s tool pricing rules.

Managed Agents add another meter. Anthropic says sessions are billed on tokens plus runtime, with session runtime at $0.08 per session-hour in the managed-agent billing table. A coding agent that runs, reads files, calls tools, and writes long outputs can outspend a short chat even when the same model is selected.

Hidden-cost range: $0 extra for web fetch beyond token costs, $10 per 1,000 web searches, and $0.08 per managed-agent session-hour before token charges.

What people pay in real use

Solo app user. A person using Claude for writing, research, and occasional coding may start with Pro or Max rather than the API. The cost tradeoff is predictability versus limits. Once paid-plan usage credits are enabled, Anthropic says follow-on usage is billed at standard API rates and appears as a separate bill line in the usage credits rules.

API developer. A developer sending long prompts with short answers can keep costs lower than a developer generating long code or reports. With Fable 5, 1 million input tokens at $10 plus 200,000 output tokens at $50 per MTok equals $20, because $10 plus 0.2 times $50 equals $20.

Coding team. Anthropic’s Claude Code cost guide says enterprise deployments average about $13 per developer per active day and $150 to $250 per developer per month, with 90% of users below $30 per active day in the Claude Code cost guide. Those figures are not Fable-only, but they show how coding usage can behave like compute.

A worked monthly total

Take a small research assistant that uses 8 million Fable 5 input tokens, 1 million output tokens, 2 million cache-read tokens, and 500 web searches in a month. Using Fable 5’s $10 input and $50 output rates from the Reuters launch report, input costs 8 times $10, or $80, and output costs 1 times $50, or $50.

Add cache and search. Cache reads at $1 per MTok make 2 million cache-read tokens cost $2, and 500 web searches at $10 per 1,000 searches cost $5. The worked total is $80 plus $50 plus $2 plus $5, or $137 before tax, discounts, marketplace handling, or unrelated seat fees.

Fable 5 vs Opus, Sonnet, Haiku, and Mythos

Fable 5 is the premium public model above the Opus tier. Anthropic’s model overview lists Fable 5 with a 1M-token context window, 128k max output, and general availability on the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry beginning June 9, 2026 in the current model overview.

The price step is clear. Fable 5 is $10 input and $50 output per MTok. Opus 4.8 is $5 and $25, Sonnet 4.6 is $3 and $15, and Haiku 4.5 is $1 and $5. Put Fable on work where quality or autonomy changes the business result. Put cheaper models on repeatable volume work.

What we verified

These checks separate app access, API billing, tool charges, and marketplace routing. That matters because a user can see a flat Claude subscription, a pay-as-you-go usage credit charge, a direct API invoice, or a cloud bill for the same family of models.

Prices can change after publication, so the safest budget is one that uses current rate cards and internal caps rather than a cached spreadsheet.

• Checked Fable 5 availability, safeguards, and the $10 and $50 API rates on Anthropic’s model launch details.
• Confirmed batch processing at $5 input and $25 output per MTok for Fable 5 through the batch processing table.
• Cross-referenced marketplace billing, CCUs at $0.01, hourly metering, monthly invoices, and AWS tax handling in the AWS marketplace section.
• Verified cancellation timing, including the 24-hour pre-renewal cancellation note and end-of-period access, through Claude subscription cancellation.

Article Highlights

• Fable 5 is priced as a premium model at $10 input and $50 output per MTok.
• App-plan bills and API bills answer different buying questions.
• Output length is the fast-moving line because output tokens cost five times input tokens.
• Cache reads and batch work can cut the API bill when the workflow fits.
• Tools, managed agents, and usage credits can move a flat plan into metered spending.

Answers to Common Questions

How much does Claude Fable 5 cost through the API?

As of June 2026, Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens before cache, batch, tool, marketplace, or discount changes.

Is Fable 5 included in Claude Pro or Max?

Anthropic is rolling Fable 5 access through subscriptions in stages, while API and consumption-based Enterprise access started at launch. Plan limits still apply.

Why can a Fable 5 bill jump so fast?

Long outputs, large files, repeated context, web search, code execution, and agent runtime all add usage. Output tokens are the biggest single meter for many writing and coding tasks.

Is Fable 5 cheaper than Opus 4.8?

No. Fable 5 is priced at twice the Opus 4.8 token rate, but it may still be rational when it reduces retries, review time, or failed agent runs.