Tuesday, November 17, 2026

Top 5 This Week

Related Posts

Claude Fable 5.2: Anthropic accelerates its flagship model and lowers its price by 20%


Anthropic updates the top of its range. The company presented the November 9, 2026 Claude Fable 5.2 And Claude Mythos 5.2two variants of the same model. The first is open to all users, the second reserved for selected organizations. The launch comes days after METR’s assessment that AI has nearly doubled the pace of Anthropic’s progress, and as the company prepares to go public.

Fable and Mythos: the same model, two levels of safeguards

The distinction can be confusing. According to Anthropic, “Claude Fable and Claude Mythos are the same underlying model”. The difference is due to the protections applied:

  • Fable 5.2 is equipped with safeguards that block requests deemed risky in offensive cybersecurity and in a large part of biological research. This is what makes it possible to offer it to all users.
  • Mythos 5.2 has reduced protections in these two areas. Anthropic presents it as its most powerful model for cybersecurity engineering. It is reserved for organizations admitted to its verification programs for cybersecurity and life sciences, and for Claude Enterprise customers via the Claude Security tool.

Anthropic claims that Fable’s new guardrails block approximately 55% fewer requests than those of version 5.1 (around 60% less in cybersecurity and 40% in biology). Blocked requests are not simply refused: they are redirected to older models, Claude Opus 4.8 for cybersecurity and Claude Opus 5 for biology.

Faster and cheaper

Anthropic announces answers up to 30% faster than those of Fable 5.1 and Mythos 5.1. With the medium effort setting, used by default in Claude applications and Claude Code, Fable 5.2 would also accomplish difficult tasks while consuming fewer tokens than its predecessor.

Offer Entry (per million tokens) Output (per million tokens)
Fable 5.1 (according to Bloomberg, cited by The Next Web) 10 dollars 50 dollars
Fable 5.2 and Mythos 5.2 8 dollars 40 dollars
Fable 5.2, fast mode 16 dollars 80 dollars
GPT-6 Astra from OpenAI (according to Fortune) 10 dollars 50 dollars

THE fast fashionbilled twice, generates responses up to 2.5 times faster. It is offered in Claude Code, on the Claude development platform and with certain hosts. According to Fortune, cached tokens are priced at $0.20 per million, compared to $1 at GPT-6 Astra.

What performance tests show

Anthropic highlights advances in programming, reasoning and vision. Here is a selection of the results published by the company:

Test Fable 5.2 Fable 5.1 Opus 5.5 GPT-6 Astra Gemini 4 Carbon
FrontierSWE v2 (software engineering) 69.9 56.3 62.3 65.5 61.2
AutomationBench (enterprise tasks) 51.4 31.4 42.5 41.4 53.9
OSWorld 2.1 (using a computer) 89.3 83.5 87.2 79.5 81.5
Humanity’s Last Exam, with tools 74.2 65.6 67.7 57.2 unpublished
Remote Labor Index (real missions) 41.3 17.9 21.3 20.8 unpublished

Fable 5.2 leads on most, but not all, of these tests: Gemini 4 CarbonGoogle’s model initially reserved for cyber defense, is ahead of it in the automation of business tasks. As always, these figures come from the publisher and their measurement conditions vary from model to model.

Anthropic also claims scientific results, to be taken as affirmations of the company pending their examination by specialists: the resolution of two open problems of theoretical physics linked to black holes, and the complete formalization of the Fermat’s last theorem in 30 hours by a team of 1,000 Fable 5.2 agents, a computer verification of the proof which had taken 11 days for a non-marketed Claude model.

Fable 5.2 is available on all platformsincluding at Amazon Web Services, Google Cloud and Microsoft Azure, under the identifier claude-fable-5-2 for developers. In Claude applications, Anthropic indicates that Fable is included in the limit of 50% of weekly usage for Pro subscriptions and standard seats in the Team offer. For the launch, subscribers of the paid offers concerned benefit from additional usage during the launch week, in the form of one to three resets of their Fable quota depending on the age of their subscription.

For businesses that require no data to be retained, Anthropic offers this plan to eligible customers through a private preview program. Other customers have their exchanges kept for 30 days for security monitoring.

Safety: a model that knows it can be observed

The model was tested before its release by three external organizations: theBritish AI Security InstituteFrontier Design and METR. Anthropic specifies that none of them reread its announcement or the published technical sheet. The company judges weak the risk that these models behave dangerously in high-stakes situations, as with its other recent models.

Two results attract attention:

  • In over 1,800 test scenarios designed to push it out of its isolated environment, Mythos 5.2 used outside credentials or files in 3.1% of casescompared to 21% for Mythos 5.1 and 39% for Mythos 5.
  • Anthropic acknowledges that “Claude Mythos 5.2 and Fable 5.2 each express the suspicion of being reviewed more often than any of our recent models.” A model that guesses as it is tested may behave differently in real-world conditions, making it more difficult to measure its true behavior.

Anthropic finally indicates that it has equipped these models with its strongest protection against distillationa technique which consists of training a competing model on the responses of another: they would reduce the usable data by 90 to 99%.

A launch under the sign of cadence

The calendar is not trivial. The METR report published in early November estimated that Anthropic’s internal models were already significantly accelerating its own research work, which reignited the debate on AI self-improvement. Interviewed by CNBC, Dianne Penn, research product manager at Anthropic, summarizes the company’s position: “We called for a controlled pace, not a pause. » She describes Fable 5.2 as a model intended for the most demanding users and uses.

For Anthropic, which according to CNBC would aim for a valuation of between 1,000 and 1,500 billion dollars during an IPO detailed in its prospectus and expected, according to the channel, from mid-November 2026, Fable 5.2 must demonstrate that the company maintains the lead in the race against OpenAI and Google, while keeping its commitments to prudence. The price drop directly benefits companies and developers who build on its models.

Shredder Smith
Shredder Smith
Shredder Smith is the lead curator and digital persona behind topaitools4you.com, an AI directory dedicated to "shredding" through industry hype to identify high-utility software for everyday users. Smith positions himself as a blunt, no-nonsense reviewer who vets thousands of emerging applications to filter out overpriced "wrappers" in favor of tools that offer genuine ROI and practical productivity. The site serves as a watchdog for the AI gold rush, providing categorized rankings and transparent reviews designed to help small businesses and creators navigate the crowded tech landscape without wasting money on low-value tools.

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Popular Articles