Claude Fable 5.2 Arrives: 20% Cheaper, and Harmless Requests Blocked About 55% Less Often

This article contains affiliate links. As an Amazon Associate I earn from qualifying purchases.

On November 9, 2026 (US time), Anthropic released Claude Fable 5.2, a new version of its top Claude model, along with Claude Mythos 5.2, the same model with some safeguards removed. Prices are 20% lower than Fable 5.1, and output is up to 30% faster. The other big change: Anthropic says harmless requests are now blocked by its safeguards about 55% less often than with Fable 5.1.

A disclosure up front: this article was written with the help of Claude, and this blog uses a paid Claude plan for research and proofreading. All performance figures below come from Anthropic’s announcement.

Claude Fable 5.2 Arrives: 20% Cheaper, and Harmless Requests Blocked About 55% Less Often

Where Fable 5.2 Fits

Claude’s current lineup runs from small to large as Haiku, Sonnet and Opus, with Fable on top. Mythos is built on the same model as Fable, with some of the safeguards for cybersecurity and biology removed, and only vetted organizations can use it.

Fable 5 came out in June and Fable 5.1 on September 1, so 5.2 arrives about two months later. Opus 5.5, released in September, was pitched as “Fable 5.1-level performance for less.” Fable 5.2 puts Fable a step above Opus 5.5 again.

Pricing and Speed

API pricing is $8 per million input tokens and $40 per million output tokens, both 20% lower than Fable 5.1’s $10 and $50. Cache reads, charged when you reuse a long context you’ve already sent, cost $0.20 per million tokens. The model ID is claude-fable-5-2.

Output is up to 30% faster than Fable 5.1. A “fast mode” speeds up generation by up to 2.5 times, at $16 per million input tokens and $80 per million output tokens, double the standard rate. Fast mode is available in Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry.

On subscriptions, Fable 5.2 is available on the Pro, Max, Ultra, Team and Enterprise plans. Anthropic also announced a limited-time arrangement letting Pro plans and standard Team seats use Fable for up to 50% of their weekly usage limits. Besides the API, it launched the same day on Amazon Web Services, Google Cloud and Microsoft Azure.

About 55% Fewer Blocks

Fable has built-in safeguards that block requests that could feed into cyberattacks or biological weapons. A blocked request gets an answer from an older Opus model instead of Fable: Opus 4.8 for cybersecurity and Opus 5 for biology, according to the announcement. That mechanism is unchanged from earlier Fable models.

The trouble is when harmless requests get blocked too. An engineer learning how to look for vulnerabilities, or a researcher checking a lab procedure, could ask an ordinary question and get a reply from the lower-tier model instead of Fable.

Anthropic says that in early testing across claude.ai, Claude Code and Cowork, Fable 5.2 was blocked about 55% less often than Fable 5.1: about 60% less in cybersecurity and about 40% less in biology. At the same time, it says the safeguards are as robust as Fable 5.1’s. The claim is that what should be blocked still is, while what needn’t be blocked is blocked less often.

Mythos 5.2 is the same model with some of those safeguards removed, used by organizations doing cybersecurity or life sciences research after an application review. In an internal test of rediscovering known vulnerabilities, Anthropic says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1.

The Benchmark Numbers

The announcement lists many benchmarks. To pick a few: on FrontierCode 2.0, which measures agentic coding, Fable 5.2 scored 49.1% (Fable 5.1: 32.4%, Opus 5.5: 45.5%, OpenAI’s GPT-6 Astra: 42.8%). On Humanity’s Last Exam with tools, it scored 74.2 (Fable 5.1: 65.6, Opus 5.5: 67.7, GPT-6 Astra: 57.2).

The biggest jump is on RoboDojo Sim, which has the model control robots in simulation: Fable 5.2 scored 64.9%, more than double Fable 5.1’s 27.6% and GPT-6 Astra’s 29.8%. On the Remote Labor Index, which measures how much of a real job gets done when it’s handed over, it rose to 41.3% from Fable 5.1’s 17.9%.

It doesn’t top every chart. On AutomationBench, which measures work automation, Google’s Gemini 4 Carbon scored 53.9 to Fable 5.2’s 51.4, and on the Elo rating of AA-Briefcase, where people grade document work, Opus 5.5’s 1800 edges out Fable 5.2’s 1785.

Anthropic also writes that during testing, Fable 5.2 solved two open problems in theoretical physics concerning black holes. Preprints have been published. The announcement notes, however, that none of the external organizations reviewed the contents of the post.

Before release, three outside organizations evaluated the model’s safety: the UK AI Security Institute, Frontier Design and METR.

What It Means for Users

The price cut and the speedup change the math on whether to use Fable for everyday work. Opus 5.5 handled work close to Fable 5.1’s level at $4 input and $20 output, 60% less than Fable 5.1. Fable 5.2 costs exactly twice Opus 5.5 and beats it on most benchmarks. That makes it easier to use Opus 5.5 day to day and switch to Fable 5.2 for the hard jobs.

Fewer blocks on harmless questions won’t show up in a benchmark table, but people who work with security or biology topics are the ones most likely to notice it. Still, the 55% figure comes from Anthropic’s early testing, and how much the day-to-day experience changes will only become clear with use.

The announcement does not say when Fable 5.1 will be retired.

Related products

You can find books and products related to this article on Amazon.

If you buy through the links above, this blog may earn a commission.

About the images

The images in this article are illustrations created with AI (ChatGPT), and you are welcome to use them. If you do, we would appreciate a link back to this article as the source.