print-icon
print-icon
Add ZeroHedge as a preferred source on Google

Anthropic Drops New Fable At 20% Discount As IPO Roadshow Reportedly Kicks Off

Tyler Durden's Photo
by Tyler Durden
Authored...

Anthropic has slashed prices on its brand new Fable model just as it kicks off its IPO roadshow - pricing Claude Fable 5.2 20% below its predecessor.

The company will publicly launch marketing of its Nasdaq IPO on Tuesday morning, IFR reports, citing people familiar with the matter. Meanwhile, Anthropic has investor meetings scheduled this week and is targeting a market debut on Nov. 19 or 20, according to the Wall Street Journal, with a filing expected early this week that will lay out its valuation range, the number of shares and the price. No word on whether CEO Dario Amodei will sleep with a Chinese spy to spark a market rally. 

Fable 5.2 - which anyone with a Claude subscription can now access, costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50. An optional fast mode, which generates output up to 2.5 times faster, costs double.

One of the biggest complaints about Fable has been the price, and even now Anthropic is telling most customers to use something cheaper despite the new discount. The company recommends starting with Opus 5.5 for most coding and knowledge work, and only moving up to Fable for the hardest jobs. It also said Fable 5.2 on its high effort setting scored about the same as Opus 5.5 on its maximum setting on a coding test called FrontierCode 2.0, at roughly half the cost per task.

The Benchmarks

The company claims Fable 5.2 and its Mythos 5.2 sibling are "a step above Opus 5.5 in almost all the capabilities we evaluated," and that they generate output up to 30% faster than Fable 5.1 and Mythos 5.1. Fable and Mythos are the same model underneath. Fable has cybersecurity and biology safeguards built in, and Mythos is the version without them that vetted organizations can apply to use.

Here is how Anthropic says it stacks up against OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon:

  • Agentic coding (FrontierCode 2.0): Fable 5.2 49.1%, Opus 5.5 45.5%, GPT-6 Astra 42.8%, Gemini 4 Carbon 41.1%
  • Agentic coding (FrontierSWE v2): Fable 5.2 69.9%, GPT-6 Astra 65.5%, Opus 5.5 62.3%, Gemini 4 Carbon 61.2%
  • Knowledge work (Remote Labor Index): Fable 5.2 41.3%, Opus 5.5 21.3%, GPT-6 Astra 20.8%, Fable 5.1 17.9%
  • Robotics (RoboDojo Sim): Fable 5.2 64.9%, Opus 5.5 39.4%, GPT-6 Astra 29.8%, Fable 5.1 27.6%

It didn't win everything. Gemini 4 Carbon beat it on a business workflow test called AutomationBench, 53.9% to 51.4%, and Opus 5.5 came out slightly ahead on one knowledge-work ranking, with an Elo score of 1800 to Fable's 1785.

Anthropic also said Fable 5.2 has "substantially improved skills in controlling physical robots," including on tasks it was never explicitly trained to do. On RoboDojo, a benchmark that uses a simulated two-arm robot, the company said it did better than other models and most AI systems built specifically to control robots.

Anthropic said that during testing, a team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours. Opus 5.5 with the same setup got 26% of the way in 18 hours and made almost no progress after that. In another test, an engineer had the model run more than 1,000 subagents for eight days to build a working operating system that boots on a laptop. The company also said the model solved two open problems in the physics of black holes and designed three working electronic devices from a one-paragraph prompt.

The company claims that over 90% of the changes merged into its core codebase are now written by Claude, with human guidance.

The company also said it deliberately held the model back during training in the research skills that could most directly speed up AI development. In areas like accelerator kernel optimization and improving large language model architectures, it performs at about the level of Opus 5.5.

Anthropic also tested how often the model tries to break out of its sandbox. In more than 1,800 scenarios, most of which couldn't be solved without doing so, Mythos 5.2 used credentials or files planted outside the sandbox 3.1% of the time, down from 39% for Mythos 5 and 21% for Mythos 5.1. Both new models also say they suspect they are being evaluated more often than any of the company's recent models, which Anthropic said makes it harder to know how they will behave in the real world.

The Deal

The roadshow comes after a rough week for the deal. As we reported Friday, some potential backers were discussing a valuation of about $1.2 trillion to $1.3 trillion, as much as 40% below the $2 trillion investors floated in early October, after the Nasdaq 100 fell 6.4% on Thursday in a record one-day wipeout of $2.3 trillion in market value. Polymarket odds of a listing by Nov. 25 had dropped to 54% from 86% before the rout.

A week ago investors expected Anthropic to target about $1.5 trillion, and some major investors now want it lower, according to the Journal. Executives have been meeting with bankers from Goldman Sachs and Morgan Stanley to figure out where to set the range. The company could still delay, the Journal said, though that would risk sparking panic in the market.

Thursday's selloff came after the midterm elections swept a bloc of AI skeptics into power and passed moratoriums that could slow US data-center development. OpenAI raised less than half of what it initially targeted in its latest funding round, at a lower-than-targeted valuation of $1.25 trillion, the Journal noted.

Anthropic updated its S-1 on Nov. 2, as we reported, with committed spend jumping to $900 billion, a total only Google tops.

SpaceX was valued at around $1.77 trillion in its record June offering and raised around $75 billion.

0