# Fable 5.1 vs 5.2: what really changes and whether to upgrade

> Fable 5.1 vs 5.2: price cut from $10/$50 to $8/$40, benchmarks, speed, fewer blocks and what breaks when you switch. Should you upgrade?

Source: Giallo Studio (giallostudio.it) · Page: https://giallostudio.it/en/blog/fable-5-2-vs-fable-5-1 · Published: 2026-11-09 · Updated: 2026-11-09

**Fable 5.1 vs 5.2: Claude Fable 5.2, released by Anthropic on 9 November 2026, costs less than Claude Fable 5.1, is faster and, in the benchmarks stated by Anthropic, scores higher in every test compared.** According to the [official pricing](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) , it drops from $10 and $50 to $8 and $40 per million tokens, for input and output: 20% less. The starting price is therefore lower, the quality is stated to be higher, and for API users there is some checking to do before changing the model ID.

The comparison starts from the [official Fable 5.2 what's-new page](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) and the [Anthropic's announcement](https://www.anthropic.com/claude-fable-and-mythos-5-2), read on 9 November 2026. For the full profile of the new model, see [Claude Fable 5.2: what it is, what changes and what it costs](https://giallostudio.it/en/blog/claude-fable-5-2-what-it-is); here we answer a single question, the one asked by anyone already on 5.1: does enough change to justify the switch?

<div class="key-points">
<p class="kp-title">In summary</p>
<ul>
<li>Fable 5.2 costs $8 and $40 per million tokens against $10 and $50 for Fable 5.1, which is 20% less on input and output, according to the <a href="https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2">official documentation</a>.</li>
<li>In the <a href="https://www.anthropic.com/claude-fable-and-mythos-5-2">announcement</a> Fable 5.2 beats Fable 5.1 in all the tests compared; among the percentage tests, the widest jump is in controlling a simulated robot (RoboDojo Sim), from 27.6% to 64.9%.</li>
<li>Output arrives up to 30% faster, and at medium effort the model completes hard tasks with fewer tokens than 5.1, according to Anthropic.</li>
<li>The safeguards on cybersecurity and biology are as robust as before, but unnecessary blocks drop by about 55% in the early tests stated by Anthropic.</li>
<li>Via the API, four rules change that can produce 400 errors: if you only use the Claude apps, you do not need to change anything.</li>
<li>Privacy unchanged: retention of at least 30 days and no zero data retention, already the case for 5.1 according to the <a href="https://support.claude.com/en/articles/15425695-covered-models">Covered Models page</a>.</li>
</ul>
</div>

## Fable 5.1 vs Fable 5.2: what are the differences? A comparison table

Fable 5.1 was released on 31 August 2026, Fable 5.2 on 9 November 2026: these are the designation dates given on [Anthropic Covered Models page](https://support.claude.com/en/articles/15425695-covered-models). Ten weeks separate the two versions. This is a summary of what is documented.

| Comparison | Fable 5.1 | Fable 5.2 |
|---|---|---|
| Exit | 31 August 2026 | 9 November 2026 |
| Input (dollars per million tokens) | 10 | 8 |
| Output (dollars per million tokens) | 50 | 40 |
| Output speed | base | [up to 30% faster](https://www.anthropic.com/claude-fable-and-mythos-5-2) |
| Tokens at medium effort | base | fewer tokens for the same task |
| Fast mode (dollars per million tokens, input / output) | not verified | 16 / 80 |
| Unnecessary safety blocks | base | about [55% fewer](https://www.anthropic.com/claude-fable-and-mythos-5-2) |
| API model ID | `claude-fable-5-1` | `claude-fable-5-2` |
| Data retention | minimum 30 days | minimum 30 days |

The first two rows are documented data; speed, tokens and blocks are Anthropic's claims about its own tests. We have not verified the fast mode of Fable 5.1, so we do not compare it.

## How much do you really save by switching to Fable 5.2?

On the price list, 20% on both items: from $10 to $8 for input and from $50 to $40 for output, as stated on the [official model page](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2). Cache reads also cost $0.20 per million tokens, keeping the same ratio of 0.025 times the input price as in earlier versions; the Batch API halves the price list to $4 and $20.

The price list, however, is not the bill. Anthropic writes that at medium effort, the default level in the Claude apps and in Claude Code, Fable 5.2 completes demanding tasks with fewer tokens than Fable 5.1. No percentage appears on the pages consulted: a further saving is plausible, but it must be measured on your own tasks. The criterion for choosing the right level is in [Fable 5.2: reasoning levels and effort](https://giallostudio.it/en/blog/fable-5-2-effort-levels), and the full price list with more calculations is in [Fable 5.2: API pricing and real costs](https://giallostudio.it/en/blog/claude-fable-5-2-api-pricing).

### An example: the communications agency in Turin

An example, declared as such: a communications agency in Turin uses Fable 5.1 via the API to prepare long, multi-step research briefs on markets and competitors. Assume 30 million input tokens and 6 million output tokens per month. On the [official pricing](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) for 5.1 comes to 30 × 10 + 6 × 50 = $600; on 5.2, 30 × 8 + 6 × 40 = $480. The saving is $120 a month, or 20%, before cache, batch and tokens saved through effort. These numbers are invented to show the method: your own tokens must be read from the invoice or the usage dashboard.

The figure an agency should watch, though, is not the saving but whether the quality of the briefs improves. That is why the test described below is better than back-of-the-envelope calculation.

## Is Fable 5.2 really better? Benchmark comparison

In the [Anthropic's announcement](https://www.anthropic.com/claude-fable-and-mythos-5-2) Fable 5.2 beats Fable 5.1 in every test where both appear. All scores are percentages unless otherwise stated, and show the maximum effort available for each model, with safeguards on.

| Test (what it measures) | Fable 5.1 | Fable 5.2 | Difference (points) |
|---|---|---|---|
| FrontierCode 2.0, Xhigh effort (code that maintainers would accept) | 32,4 | 49,1 | +16,7 |
| FrontierSWE v2 (software engineering) | 56,3 | 69,9 | +13,6 |
| AutomationBench (business workflows) | 31,4 | 51,4 | +20,0 |
| Terminal-Bench-Science 0.1 (scientific research with agents) | 52,6 | 74,6 | +22,0 |
| WANDR, relaxed score (data collection with agents) | 68,7 | 82,1 | +13,4 |
| OSWorld 2.1, partial offline set (computer use) | 83,5 | 89,3 | +5,8 |
| RoboDojo Sim (control of a simulated robot) | 27,6 | 64,9 | +37,3 |
| Remote Labor Index, Fable 5.2 at High effort (remote jobs assigned to an agent) | 17,9 | 41,3 | +23,4 |
| Humanity's Last Exam, with tools (expert knowledge) | 65,6 | 74,2 | +8,6 |
| Chartography, without tools (chart reading) | 44,8 | 73,4 | +28,6 |

Three cautious readings:

- **The numbers are stated by Anthropic.** These are the results of its own table, not independent checks, and on the pages consulted on 9 November 2026 we found no replications by third parties. For AA-Briefcase, a knowledge-work test scored against a rubric, the score rises from 61.5 to 66.1: a smaller step.
- **The gain is not uniform.** The tests closest to office work (AutomationBench, Remote Labor Index, Chartography) and to research with agents improve a lot. Computer use (OSWorld) improves less, probably because it started out high.
- **Some tests are internal.** Anthropic specifies that RoboDojo Sim uses maximum effort and was run internally, so the +37.3 is the number to read with the most caution.

The full picture, with the columns for Opus 5.5 and the competitors, is in [Claude Fable 5.2 benchmarks explained](https://giallostudio.it/en/blog/claude-fable-5-2-benchmarks). The comparison with the lower-tier model is in [Fable 5.2 vs Opus 5.5](https://giallostudio.it/en/blog/fable-5-2-vs-opus-5-5). The Mythos version, which lacks the safeguards on cybersecurity and biology, has a table of its own: [Claude Mythos 5.2: what it is and who can use it](https://giallostudio.it/en/blog/claude-mythos-5-2-what-it-is).

## Speed and tokens: what gets better in everyday use?

According to the [announcement](https://www.anthropic.com/claude-fable-and-mythos-5-2), Fable 5.2 generates output up to 30% faster than Fable 5.1. “Up to” means a ceiling, not an average: on short requests the difference may not show. For those in a hurry there is fast mode, up to 2.5 times faster at $16 per million input tokens and $80 per million output tokens; that is double the standard price, so use it only where latency has value.

More interesting is efficiency. On the FrontierCode 2.0 test, the [announcement](https://www.anthropic.com/claude-fable-and-mythos-5-2) states that Fable 5.2 at High effort scores almost the same as Opus 5.5 at Max effort (45.3% against 45.5%), for about half the cost per task. It is a comparison with Opus, not with 5.1, but it says that moderate effort is enough more often than before.

### Two new behaviors to know about

The [what's-new page](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) flags two differences that you can see without touching the code:

- **The model thinks even when told not to.** Fable 5.1 usually respected a line such as “reply directly without reasoning”. In Anthropic's tests, at medium effort or higher Fable 5.2 thinks anyway. For simple requests you need to lower the effort, not rewrite the prompt.
- **Sometimes it stops before finishing.** On long, multi-step tasks it may end its turn with a list of what remains, more often at low and medium effort. Ask for an explicit completion criterion.

The rest of the known limitations is in [Fable 5.2 known limitations and how to work around them](https://giallostudio.it/en/blog/claude-fable-5-2-limitations).

## Why do fewer safety blocks matter for an SME?

Fable 5.2's safeguards on cybersecurity and biology are as robust as those of Fable 5.1, Anthropic writes, but they produce fewer unnecessary blocks. In early tests on claude.ai, Claude Code and Cowork, the company observed about 55% fewer blocks than Fable 5.1: about 60% fewer in cybersecurity and 40% in biology, according to the [official announcement](https://www.anthropic.com/claude-fable-and-mythos-5-2).

In practice this means fewer diverted responses. When a request is blocked, the response comes from Opus 4.8 for cybersecurity and from Opus 5 for biology. For legitimate requests close to sensitive topics, such as an IT manager analyzing a suspicious log, one block fewer means one interruption fewer. These are still Anthropic's figures on an initial sample: the true frequency on your own cases only shows up when you use it.

## What breaks when you switch via the API (and what does not)?

If your requests are built by the Claude apps, Claude Code, Managed Agents or the Agent SDK, the change to history handling does not concern you. If instead your own code builds them, the [official documentation](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) lists four changes that can produce 400 errors:

<div class="callout"><strong>The four changes, in two lines.</strong> Editing the earlier turns of a conversation invalidates the reasoning blocks and returns 400, on every account; the old tool <code>computer_20251124</code> returns 400 on every platform except Amazon Bedrock. In addition, Fable 5.2's reasoning blocks can only be read by Fable 5.2 and Mythos 5.2, and the advisor of a Fable 5.2 executor must be Fable 5.2 or Mythos 5.2. The step-by-step path is in <a href="https://giallostudio.it/en/blog/fable-5-2-api-migration">Migrating to Fable 5.2: API changes and 400 errors</a>.</div>

Changing the ID is the easy part: from `claude-fable-5-1` a `claude-fable-5-2`. If you are coming from Fable 5, with adaptive thinking already always on, the documentation lists as the only breaking change the forced tool choice (`tool_choice` of type any or tool), which returns 400. For those coming from 5.1, that rule and always-on thinking are nothing new.

## Should you upgrade to Fable 5.2 right away?

The answer depends on how you use it.

<div class="compare">
<div class="compare-yes"><h4>Switch now</h4><ul>
<li>You use Fable from the Claude apps: just change the model in the selector.</li>
<li>You have continuous API usage and a test environment where you can try the code.</li>
<li>Your tasks are long and multi-step, and 5.1 sometimes does not finish them.</li>
<li>Monthly cost matters: the lower price list pays off immediately.</li>
</ul></div>
<div><h4>Wait a few days</h4><ul>
<li>Your code edits or rebuilds the earlier turns of the conversation.</li>
<li>You use the computer tool <code>computer_20251124</code> outside Amazon Bedrock.</li>
<li>You have prompts written for 5.1 that ask it to “not reason” and you have not retested them.</li>
<li>You have flows that depend on data under zero data retention: 5.1 and 5.2 are both Covered Models, so changing model does not solve the problem.</li>
</ul></div>
</div>

There is an extra incentive for the apps. According to [Anthropic's launch offers page](https://support.claude.com/en/articles/17506494), from 9 to 16 November 2026 the Pro plans and the standard seats of the Team plan have all Fable models included up to 50% of the weekly limits, plus a limits reset. The offer also covers Fable 5.1, so in that week you can put the two models side by side at no extra cost. Free plans are excluded. The detail by plan is in [Fable 5.2 on Pro, Max and Team: limits and offer](https://giallostudio.it/en/blog/claude-fable-5-2-plans-usage-limits).

If you are still choosing the family and not the version upgrade, we point you to the September comparison [Opus 5 vs Fable 5.1](https://giallostudio.it/en/blog/opus-5-vs-fable-5-1), to [Fable 5.1 vs Sonnet 5](https://giallostudio.it/en/blog/fable-5-1-vs-sonnet-5) and [GPT-6 Astra vs Fable 5.1](https://giallostudio.it/en/blog/gpt-6-astra-vs-fable-5-1): they were written about the previous generation, but the method of choosing is the same. The current price lists for the whole family are in [Claude model pricing: API rates and the cost of getting work done](https://giallostudio.it/en/blog/claude-api-pricing).

## Mistakes to avoid when upgrading

- **Changing the ID in production without testing.** A 400 error on editing history may only appear after the switch, even if you never saw it with 5.1.
- **Reading the benchmarks as a guarantee.** They are Anthropic's tests on generic tasks. Your case may improve less, or more.
- **Carrying 5.1 prompts over as they are.** Instructions to “not reason” or to use short sentences produce different effects: 5.2 follows style rules to the letter, and a rigid limit on sentence length can chop the text into fragments.
- **Keeping effort at maximum out of habit.** The default level in the API is high and the documentation recommends starting there; max is mainly for measuring the model's ceiling, or for the rare tasks where time and tokens do not matter.
- **Thinking the switch fixes privacy.** The data regime is identical, so a flow that was not suitable for 5.1 does not become suitable with 5.2.

## How to upgrade in four steps

<ol class="steps">
<li><strong>Pick 15-20 real requests</strong> already handled with Fable 5.1, for which you have a good answer, anonymized.</li>
<li><strong>Rerun them on Fable 5.2 at high effort</strong>, leaving the rest of the prompt unchanged, and note quality, time and tokens.</li>
<li><strong>If you use the API, run the history and tool tests</strong> in the test environment: turn editing, computer use, advisor.</li>
<li><strong>Move the least critical traffic first</strong>, check the real cost for a week and only then move the rest.</li>
</ol>

## Upgrade, yes, but with a test

Moving from Fable 5.1 to Fable 5.2 is rare in its simplicity: it costs less, is stated to be more capable and faster, and privacy does not change. The only real cost falls on those who build on the API, and it is paid once, in a few hours of testing. For the usage of an agency or a mid-sized firm, the decision takes an afternoon, not a month.

If you want someone to prepare those 15-20 cases, measure them with you and tell you whether your tasks need Fable, Opus 5.5 or a smaller model, [Giallo Studio's AI consultancy](https://giallostudio.it/en/ai-consulting) starts from a test on your own data, before any migration. Before deciding, review the [official Anthropic pages](https://platform.claude.com/docs/en/models/fable-5-2/overview): versions, prices and offers change quickly.

## Frequently asked questions

### What is the difference between Fable 5.1 and Fable 5.2?

Fable 5.2, released on 9 November 2026, costs 20% less than Fable 5.1 ($8 and $40 per million tokens versus $10 and $50), generates output up to 30% faster and, according to the benchmarks stated by Anthropic, scores higher in every test compared. According to Anthropic it also blocks about 55% fewer requests for safety reasons. In return, it introduces four API changes that can break code written for 5.1.

### Is Fable 5.2 cheaper than Fable 5.1?

Yes, on both input and output. According to the official documentation consulted on 9 November 2026, Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, while Fable 5.1 cost $10 and $50. The saving on the price list is 20% on both items; the real bill also depends on the tokens consumed.

### Should you switch from Fable 5.1 to Fable 5.2 right away?

In the Claude apps, yes: you only change the choice in the model selector. Via the API, switch after testing your code in a test environment, because the rules change for editing earlier turns, for computer use and for the advisor. The lower price list pays back the testing quickly if usage is continuous.

### Is Fable 5.2 faster than Fable 5.1?

Yes. Anthropic states that Fable 5.2 and Mythos 5.2 generate output up to 30% faster than Fable 5.1 and Mythos 5.1. There is also an optional fast mode, up to 2.5 times faster, at $16 and $80 per million tokens.

### What breaks when moving from Fable 5.1 to Fable 5.2 via the API?

Four things, according to the official documentation. Editing earlier turns of a conversation returns a 400 error, the old computer_20251124 tool returns 400 except on Amazon Bedrock, Fable 5.2's reasoning blocks can only be read by Fable 5.2 and Mythos 5.2, and the advisor of a Fable 5.2 executor must be Fable 5.2 or Mythos 5.2.

### Is Fable 5.1 still available after the release of Fable 5.2?

As of 9 November 2026, Anthropic's Covered Models page still lists Fable 5.1 as generally available. No retirement date for Fable 5.1 appears on the official pages consulted: before planning, check Anthropic's deprecations page.

### Does Fable 5.2 change anything on privacy compared with Fable 5.1?

No, the regime is the same. Fable 5.1 is also a Covered Model, with a minimum retention of 30 days for prompts and responses and no zero data retention, unless Anthropic expressly authorizes it. If you had already assessed 5.1 on this point, there is no new constraint, but you still need to review the data you send.
