# Claude Fable 5.2 Pricing: API Rates and Real Costs

> Claude Fable 5.2 pricing: $8 input and $40 output per million tokens. Cache, batch, fast mode and a step-by-step cost example for a real business.

Source: Giallo Studio (giallostudio.it) · Page: https://giallostudio.it/en/blog/claude-fable-5-2-api-pricing · Published: 2026-11-09 · Updated: 2026-11-09

**Claude Fable 5.2, released by Anthropic on 9 November 2026, is priced at $8 per million input tokens and $40 per million output tokens.** That is 20% less than Fable 5.1 and double Claude Opus 5.5. This comes from the [official model page](https://platform.claude.com/docs/en/models/fable-5-2/overview), consulted on 9 November 2026. The price list has few lines, but the real bill depends on things the list does not show: how much the model reasons, how much of the prompt repeats, and whether the answer is needed immediately or can wait.

This article lines up the items (input, output, cache, batch, fast mode), compares them with the other Claude models and works through a transparent calculation for an example case. To learn what the model is and where it sits in the family, see the guide [Claude Fable 5.2: what it is, what changes, what it costs](https://giallostudio.it/en/blog/claude-fable-5-2-what-it-is): here we only talk about money.

<div class="key-points">
<p class="kp-title">In summary</p>
<ul>
<li>Claude Fable 5.2 costs <a href="https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2">$8 per million input tokens and $40 per million output tokens</a>, against $10 and $50 for Fable 5.1.</li>
<li>Cache reads cost <a href="https://platform.claude.com/docs/en/models/fable-5-2/overview">$0.20 per million</a>, one fortieth of the base input price; cache writes cost $10 (5 minutes) or $16 (1 hour).</li>
<li>The Batch API halves the bill: <a href="https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2">$4 and $20</a>; fast mode doubles it, at $16 and $80, for generation up to 2.5 times faster.</li>
<li>Per token, Fable 5.2 costs twice as much as <a href="https://platform.claude.com/docs/en/models/fable-5-2/overview">Opus 5.5 ($4 and $20)</a> and Anthropic recommends starting from Opus.</li>
<li>In our support-center example, 1,000 conversations cost $80 on Fable 5.2, $40 on Opus 5.5 and $20 on Sonnet 5.5.</li>
<li>The price list is in dollars, with no euro prices on the official pages consulted on 9 November 2026: for Italy, VAT is added on top.</li>
</ul>
</div>

## How much does Claude Fable 5.2 cost per million tokens?

The full price list, in dollars per million tokens, according to the [“What's new” page of the documentation](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) and the [model card](https://platform.claude.com/docs/en/models/fable-5-2/overview):

| Price item | Fable 5.2 ($ per million tokens) |
|---|---|
| Base input | 8 |
| Output | 40 |
| Cache write, 5 minutes | 10 |
| Cache write, 1 hour | 16 |
| Cache read | 0,20 |
| Batch API, input | 4 |
| Batch API, output | 20 |
| Fast mode, input | 16 |
| Fast mode, output | 80 |

Mythos 5.2, which is the same model without the safeguards on cybersecurity and biology, [has the same prices](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2). Price is therefore not a reason to choose one or the other: this is explained in [Fable 5.2 vs Mythos 5.2](https://giallostudio.it/en/blog/fable-5-2-vs-mythos-5-2).

The official page points to a separate page for prices tied to data residency and tools (for example web search): we do not cover them here and have not verified them. If your project uses them, check that page before preparing the quote.

### How do you read the price list? Input only counts in part

Cost splits in two: what you send to the model (input) and what the model writes (output). Output costs five times as much as input. On the output side, moreover, [Fable 5.2 always reasons](https://platform.claude.com/docs/en/models/fable-5-2/overview): adaptive thinking cannot be turned off, and depth is set with the effort parameter, which the documentation describes as a control of depth, latency and cost. Anyone who reads only the price per token risks underestimating the most expensive item. To understand what a token is and why the count depends on language and type of text, see [AI tokens: what they are and why they matter](https://giallostudio.it/en/blog/what-are-ai-tokens).

One detail for those comparing with older models: Fable 5.2 uses the [tokenizer introduced with Claude Opus 4.7](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2). The same text can therefore correspond to a different number of tokens than on an earlier model. An honest comparison is made on the cost of a finished task, not on price per token alone.

## How does the price compare with Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5?

The [official page](https://platform.claude.com/docs/en/models/fable-5-2/overview) compares the current family. The Fable 5.1 price comes from the [what's-new page](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2).

| Model | Input ($ per million) | Output ($ per million) | Ratio to Fable 5.2 |
|---|---|---|---|
| **Fable 5.2** | 8 | 40 | 1 |
| Fable 5.1 | 10 | 50 | 1.25 times more expensive |
| Opus 5.5 | 4 | 20 | half |
| Sonnet 5.5 | 2 | 10 | a quarter |
| Haiku 5.5 | from 0.10 | from 0.50 | about one eightieth |

The “from” in Haiku 5.5's price comes from the official page: the price tier changes with prompt length, as we explain in [Claude model pricing: API rates and the cost of getting work done](https://giallostudio.it/en/blog/claude-api-pricing). A low price does not mean the same quality: Haiku does classification and extraction, not jobs that require hours of reasoning.

Three practical conclusions:

- **Fable 5.2 is not the model to put everywhere.** Anthropic itself, in the [what's-new page](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2), recommends starting from the latest Opus and moving to Fable when tests on Opus at higher effort are still not enough. The choice is covered in detail in [Fable 5.2 vs Opus 5.5](https://giallostudio.it/en/blog/fable-5-2-vs-opus-5-5).
- **Compared with Fable 5.1, the saving is real.** Not only because of the lower list price: according to the [announcement](https://www.anthropic.com/claude-fable-and-mythos-5-2), at medium effort the new version completes demanding tasks with fewer tokens. The comparison is in [Fable 5.2 vs Fable 5.1](https://giallostudio.it/en/blog/fable-5-2-vs-fable-5-1).
- **The price of an old model is not a reference point.** For details on Opus 5.5, see [Claude Opus 5.5: benchmarks, prices and comparison](https://giallostudio.it/en/blog/claude-opus-5-5-benchmarks-pricing).

## How much does it really cost? An example: the support center of an air-conditioning company

An example, with invented but reasonable numbers, which you can replace with your own. An air-conditioning company uses an AI assistant to answer customers about faults, warranty and maintenance. Assumptions:

- 1,000 conversations a month;
- each conversation sends the model 6,000 input tokens (instructions, product sheet, history) and receives 800 output tokens, reasoning included;
- no cache, no batch.

The calculation, on the [official pricing](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2), is simple: 1,000 × 6,000 = 6 million input tokens and 1,000 × 800 = 0.8 million output tokens.

| Model (prices from the [price list](https://platform.claude.com/docs/en/models/fable-5-2/overview)) | Input: 6 million × price ($) | Output: 0.8 million × price ($) | Monthly total ($) |
|---|---|---|---|
| Fable 5.2 | 48 | 32 | 80 |
| Fable 5.1 | 60 | 40 | 100 |
| Opus 5.5 | 24 | 16 | 40 |
| Sonnet 5.5 | 12 | 8 | 20 |
| Haiku 5.5 (base price) | 0,60 | 0,40 | 1,00 |

On Fable 5.2 each conversation costs $0.08 (6,000 × 8 + 800 × 40, divided by 1 million), eight cents at the [official pricing](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2). These are small figures, and that is the point: at a low volume, the difference between Fable and Opus ($40 a month) is irrelevant next to an operator's time. At a volume a hundred times higher it becomes $4,000 a month. The right question is not “how much does the token cost?” but “does Fable avoid enough rework to pay for the difference?”. For a question about an air conditioner's manual, almost never. For a diagnosis on an industrial plant with twenty documents to cross-check, maybe.

### Cache: how a repeated prompt costs one fortieth

In that 6,000-token prompt, 4,000 tokens are the same in every conversation: instructions and product sheet. The [prompt caching](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) lets you write that part once and re-read it at $0.20 per million instead of $8. Let's assume the 5-minute cache is rewritten 100 times in the month and read the other 900 times:

| Cost item (prices from the [price list](https://platform.claude.com/docs/en/models/fable-5-2/overview)) | Calculating | Cost ($) |
|---|---|---|
| [Variable Part](https://platform.claude.com/docs/en/models/fable-5-2/overview), 2,000 tokens × 1,000 conversations | 2 million × 8 | 16 |
| [Cache write](https://platform.claude.com/docs/en/models/fable-5-2/overview), 4,000 tokens × 100 times | 0.4 million × 10 | 4 |
| [Cache read](https://platform.claude.com/docs/en/models/fable-5-2/overview), 4,000 tokens × 900 times | 3.6 million × 0.20 | 0,72 |
| [Output](https://platform.claude.com/docs/en/models/fable-5-2/overview) | 0.8 million × 40 | 32 |
| **Monthly Total** | | **52,72** |

From $80 to about $53 ([cache price list](https://platform.claude.com/docs/en/models/fable-5-2/overview)): 34% less, without changing either the model or the quality. The saving depends on how long the cache stays valid; with intermittent traffic, the 1-hour write ($16) can cost less than repeated 5-minute writes. The mechanism, with Sonnet 5.5 prices, is explained in [Sonnet 5.5 and prompt caching: prices and savings](https://giallostudio.it/en/blog/claude-sonnet-5-5-prompt-caching-pricing). In the example we did not combine cache and batch, because the pages consulted do not say how they add up.

## How can you lower your Fable 5.2 bill? Five levers

<ol class="steps">
<li><strong>Pick the model per task.</strong> Simple questions to Sonnet 5.5 or Haiku 5.5, the rest to Opus 5.5, and Fable only for the cases that deserve it. Anthropic describes this scheme in the <a href="https://www.anthropic.com/claude-fable-and-mythos-5-2">announcement</a>.</li>
<li><strong>Cache what repeats.</strong> Instructions, product sheets, rules: they are the fixed part of the prompt. Reads cost $0.20 instead of $8, according to the <a href="https://platform.claude.com/docs/en/models/fable-5-2/overview">price list</a>.</li>
<li><strong>Use the Batch API for what can wait.</strong> Classifying a ticket archive, reviewing contracts in a queue, preparing overnight reports: <a href="https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2">$4 and $20</a>.</li>
<li><strong>Tune the effort.</strong> It is the most underrated lever. According to the <a href="https://www.anthropic.com/claude-fable-and-mythos-5-2">announcement</a>, on FrontierCode 2.0 Fable 5.2 at High effort reaches 45.3% against Opus 5.5's 45.5% at Max, for about half the cost per task: a figure declared by Anthropic. Details in <a href="https://giallostudio.it/en/blog/fable-5-2-effort-levels">Fable 5.2 effort levels</a>.</li>
<li><strong>Combine models of different tiers.</strong> Fable 5.2 can act as an Advisor consulted by smaller models, or as an Orchestrator that hands simple tasks to cheaper models, as the <a href="https://www.anthropic.com/claude-fable-and-mythos-5-2">announcement</a>says. How it works is covered in <a href="https://giallostudio.it/en/blog/fable-5-2-advisor-orchestrator">Fable 5.2 Advisor and Orchestrator</a>.</li>
</ol>

## When is fast mode worth double, and when is it not?

Fast mode costs [$16 and $80](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) and speeds up generation by up to 2.5 times. In the example above it would take the bill from $80 to $160 a month.

<div class="compare">
<div class="compare-yes"><h4>When it makes sense</h4><ul>
<li>A person is waiting for the answer in real time, for example an operator consulting the assistant during a phone call.</li>
<li>The time saved is worth more than the surcharge, as in an interactive development session.</li>
<li>The task is short and the extra cost is a few cents.</li>
</ul></div>
<div><h4>When to Avoid It</h4><ul>
<li>Overnight processing, batches, reports: nobody is waiting, so you pay double for nothing.</li>
<li>Agents that work for hours: speed matters less than total cost.</li>
<li>You have not measured the real waiting time first: the bottleneck is often elsewhere.</li>
</ul></div>
</div>

Where is fast mode available? According to the [documentation](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2), on Messages API requests on the Claude API, on Claude Models on Google Cloud and on Microsoft Foundry, where it appears as a separate model `claude-fable-5-2-fast`. On the other platforms, check the official page.

## What is not included in the price per token?

Some items do not appear in the price list but weigh on the bill:

- **Subagents.** According to the [what's-new page](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2), in a product that invites delegation, the model delegates readily, and subagents can add a lot to the cost of a session.
- **Unrequested reasoning.** At medium effort or higher the model reasons even when told not to: the same page lists it among the known behaviors. Other limits are covered in [Fable 5.2: the limits Anthropic declares](https://giallostudio.it/en/blog/claude-fable-5-2-limitations).
- **Refusals.** The model can refuse a request, and the documentation indicates that if the refusal comes before any output, billing depends on the category. Falling back to another model refunds the cache cost of the switch.
- **Data retention.** It is not a cost in euros but a constraint: Fable 5.2 requires 30 days of retention, and zero data retention (ZDR) is not available without authorization. We discuss it in [privacy and data retention](https://giallostudio.it/en/blog/fable-5-2-privacy-data-retention).

Another useful distinction: the API price is not the subscription price. Anyone using Claude in the browser pays for a plan with usage limits, not for tokens; the support page on the [launch offers](https://support.claude.com/en/articles/17506494) explains what is included per plan, and we summarize it in [Fable 5.2 in Claude plans](https://giallostudio.it/en/blog/claude-fable-5-2-plans-usage-limits). For the relationship between Team plans and API credits, see [Claude Max, Team and API credits](https://giallostudio.it/en/blog/claude-max-team-api-credits).

## What mistakes should you avoid when preparing a quote?

1. **Counting only the input tokens.** Output costs five times as much and includes reasoning: estimating it by eye throws off the quote more than any discount.
2. **Comparing Fable's price with that of a model on a different tokenizer.** The number of tokens changes: compare the cost of the finished task.
3. **Measuring cost on a single case.** Take 15-20 real requests, with their real lengths, and run Opus 5.5 and Fable 5.2 at the same effort. The average cost is worth more than any price list.
4. **Forgetting VAT and exchange rates.** The price list is in dollars, Italian VAT is not included, and no euro prices appear on the pages consulted on 9 November 2026. Build the budget with a prudent exchange rate.
5. **Switching to Fable “to be safe”.** With Opus 5.5 at half the price you often get the same result. The general method for calculating costs is in [AI API costs: how to calculate them](https://giallostudio.it/en/blog/how-to-calculate-ai-api-costs).

## How do you estimate your own bill in three steps?

A method we use with businesses, with no need to be a developer:

<ul class="checklist">
<li>Collect 15-20 real requests, of different lengths and difficulty, and note the input and output tokens the API returns for each.</li>
<li>Run the same cases on Sonnet 5.5, Opus 5.5 and Fable 5.2, at the same effort, and compare quality before cost.</li>
<li>Multiply the average cost by the monthly volume, then apply cache and batch where it makes sense and calculate again.</li>
<li>Add a margin for subagents and for reasoning: run a one-week trial before committing the annual budget.</li>
</ul>

If the volume is modest, the cost of Fable 5.2 is not the problem: the problem is the time of whoever sets up and monitors the system. If the volume is high, the problem is architecture: which model does what, how much of the prompt can be cached, what can go in a queue. For models in business more generally, there is also [Fable 5.2 for business: when you really need it](https://giallostudio.it/en/blog/claude-fable-5-2-for-business).

If you want a quote built on your volumes rather than on an example, [Giallo Studio](https://giallostudio.it/en/pricing) prepares it with you: we start from your processes, measure the cost per task across several models and tell you whether Fable 5.2 justifies the spend. To see how we work, there is the page [AI Consulting](https://giallostudio.it/en/ai-consulting).

## Frequently asked questions

### How much does Claude Fable 5.2 cost via the API?

It costs $8 per million input tokens and $40 per million output tokens, according to the official page consulted on 9 November 2026. Cache writes cost $10 (5 minutes) or $16 (1 hour) per million, and cache reads cost $0.20. Prices are in dollars: for Italy, VAT is not included.

### Is Fable 5.2 cheaper than Fable 5.1?

Yes, by 20%. Fable 5.1 cost $10 for input and $50 for output per million tokens; Fable 5.2 costs $8 and $40. Anthropic also states that at medium effort Fable 5.2 completes demanding tasks with fewer tokens than 5.1, so the real saving may be higher: measure it on your own cases.

### How much does Fable 5.2 fast mode cost?

$16 per million input tokens and $80 per million output tokens, that is, double the standard price list. In return, generation is up to 2.5 times faster. It only makes sense where a person is waiting for the answer in real time; for queued processing it is a waste.

### Does the Batch API apply to Fable 5.2 too?

Yes: the official documentation indicates a 50% discount on input and output, so $4 and $20 per million tokens. It works for jobs that do not need an immediate answer, such as classifying a ticket archive overnight.

### Is Fable 5.2 worth it over Opus 5.5 on price?

Per token it costs twice as much as Opus 5.5 ($4 and $20). It pays off only if it solves tasks that Opus 5.5 does not close well, even at high effort. Anthropic states that on FrontierCode Fable 5.2 at High effort reaches about the score of Opus 5.5 at Max for about half the cost per task: this is a vendor-declared figure, not verified by third parties.

### Are there euro prices for Fable 5.2?

No. No euro prices appear on the official pages consulted on 9 November 2026: Anthropic publishes its price list in dollars. For a quote in euros, convert at the day's exchange rate and add VAT. Anyone buying through Amazon Bedrock, Google Cloud or Microsoft Foundry should check the price list and billing of the chosen platform.
