Claude Fable 5.2: what changed, what it costs and when to use it

Anthropic released Claude Fable 5.2 on 9 November 2026, alongside Claude Mythos 5.2. It is cheaper than Fable 5.1, Anthropic says it is faster, and it is aimed at long, demanding work. The real question for a business is whether it is worth paying for on your own tasks.

Here is what changed, what it costs and how to test it, based on Anthropic's announcement and documentation.

What changed from Fable 5.1

Anthropic positions Fable 5.2 for hard reasoning and long-running agent work. What changed:

• Lower prices. Input and output now cost $8 and $40 per million tokens, down from $10 and $50, a 20% cut.
• Faster output. Anthropic says it generates outputs up to 30% faster than Fable 5.1.
• Leaner at medium effort. Anthropic says it can complete challenging tasks with fewer tokens than Fable 5.1.
• Fewer blocks. Anthropic reports about 55% fewer blocked requests than Fable 5.1 in early testing.

It has a 1M-token context window and up to 128K output tokens, reads text and images, and has a June 2026 knowledge cutoff. It is the same model as Mythos 5.2, with extra biology and cybersecurity safeguards so it can be offered for general use. Fable 5.1 is not deprecated, so you can move on your own timeline.

What it costs, with a worked example

Per million tokens, Fable 5.2 costs $8 input, $40 output and $0.20 for cache reads. The Batch API halves input and output to $4 and $20. For comparison, Opus 5.5 costs $4/$20, Sonnet 5.5 $2/$10 and Haiku 5.5 from $0.10/$0.50, so Fable 5.2 is twice the per-token price of Opus 5.5. There is no long-context surcharge.

Fast mode is separate from the speed gain above. It runs the same model with no change in intelligence and gives up to 2.5x more output tokens per second, at $16 input and $80 output. It does not shorten the wait for the first token or work with the Batch API.

For example, say a monthly research-and-reporting workflow sends 20 million input tokens and receives 2 million output tokens:

In this example, moving from Fable 5.1 saves $60 a month, and Opus 5.5 handles the same volume for half the price. But price per token is not cost per task. Anthropic says that on FrontierCode 2.0, a coding benchmark, Fable 5.2 at high effort roughly matched Opus 5.5's best score for about half the cost per task ($6.72 against $12.82).

Where to get Fable 5.2

• Claude apps. Not on Free. On Pro, Fable is paid with usage credits. Max ($100/month) and Ultra ($200/month) include it up to 50% of weekly limits, as do Team premium seats, and Enterprise includes it. Anthropic warns that Fable models use up limits faster.
• Launch offer. Until 16 November 2026 at 11:59 pm PT, eligible Pro plans and standard Team seats get Fable included up to 50% of weekly limits, plus one Fable reset.
• Claude Code. Supported from v2.1.321. Pick it with /model fable or claude --model fable, because Fable is not the default anywhere.
• The API. The model ID is claude-fable-5-2 (see the model overview).
• Cloud platforms. The same ID works on AWS through Claude Platform on AWS, on Google Cloud through Claude Models on Google Cloud (Anthropic or Google managed) and on Microsoft Foundry on Anthropic infrastructure. Fast mode is not listed for AWS.

Fable 5.2 or a smaller Claude model?

Anthropic's own advice matters most: start with Claude Opus 5.5, which it calls "our strongest daily driver", and move up only when the work needs it. Its docs say to switch when your evaluations of Opus at higher effort still fall short.

Pick Fable 5.2 when an agent must work for hours across many steps, when research has to end in a finished document, spreadsheet or deck, or when Opus 5.5 at high effort still misses your quality bar.

Pick a smaller model when the work is routine and high-volume, such as tagging tickets or answering FAQs, where Sonnet 5.5 or Haiku 5.5 is usually the better buy. Anthropic calls Fable 5.2 a step above Opus 5.5 in almost all the capabilities it evaluated, yet its own system card shows Opus 5.5 ahead on two ratings, GDPval-AA and AA-Briefcase.

You can also mix models: Haiku, Sonnet or Opus can consult Fable 5.2 as an advisor, or Fable 5.2 can act as an orchestrator that delegates simpler tasks. Moving a conversation from Fable 5.2 to another model loses its reasoning.

Test it on your own work before switching

1. Pick 20 to 50 real tasks with known good answers.
2. Set a baseline with your current model, for example Opus 5.5 at high effort.
3. Run Fable 5.2 at high effort. Anthropic says most of the capability gain is already present at high. Defaults differ: high on the API, medium in Claude Code and the Claude apps.
4. Score cost per task, not per token, including subagents, which can make up a large share of a session's cost.
5. Check for unfinished work. On long tasks the model sometimes ends a turn early, more often at low and medium effort.
6. Switch only the tasks where it clearly wins.

Caveats worth taking seriously

Private data. Fable 5.2 requires 30-day data retention and is not available under zero data retention unless Anthropic expressly authorizes it. Decide which customer or financial data should reach it at all.

Safeguards. Classifiers block high-risk cybersecurity activity and most research-biology tasks. In Claude's apps and Claude Code, Opus 4.8 or Opus 5 answers blocked requests instead. On the API, a declined request returns stop_reason "refusal" and fallback is opt-in, so your code must handle it.

Outputs. Review numbers, sources and anything customer-facing before it ships. Anthropic's single-turn safety evaluations covered seven languages, including Arabic, so test in the languages your customers use.

Developers. From Fable 5.1 the move is mostly drop-in: change the model ID, then check the four breaking changes in Anthropic's migration guide. For example, the older computer_20251124 tool now returns an error, so use computer_toolset_20260801. Code coming from other models must also drop temperature, top_p and top_k, and stop prefilling the assistant message.

How ADMOV can help

ADMOV's LLM setup and integration service is built for this decision. From Istanbul, we help businesses in the Gulf, Türkiye and worldwide test Fable 5.2 against cheaper Claude models on real tasks, route only the hard work to it, and add cost limits, guardrails, refusal handling and data checks.

If your team wants these skills in-house, Admov Academy offers practical AI courses, a prompt bank and certificates.

Book a free call to plan your switch to Fable 5.2, or to confirm you don't need it yet.