Anthropic Releases Claude Fable 5.2 and Mythos 5.2 on the Eve of Its IPO

Download the app

+1 Anthropic releases two new models on the eve of its listing; Nvidia and six other giants plan to invest $26 billion in a private placement. The Paper reporter Hu Hanyan, 2026-11-10 13:49 Source: The Paper ∙ 10% Company > Listen to the full text Font size

As the AI (artificial intelligence) sector has recently experienced sharp volatility, Anthropic is still accelerating its IPO (initial public offering) and has launched two new-generation frontier models on the eve of its listing.

On November 9 local time, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2. This is the first time Anthropic has launched new frontier models since the company's CEO Dario Amodei called in September this year for slowing the pace of AI research and development.

At the same time, seven tech giants including Nvidia and Google plan to invest $26 billion in Anthropic through a private placement to support this highly anticipated IPO.

Seven tech giants plan to invest $26 billion; IPO terms still subject to change

According to foreign media reports, seven tech giants—Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC—will invest a total of $26 billion in Anthropic through a private placement at the IPO issue price.

This lineup of investors covers almost every core link in the current AI industry chain, spanning cloud computing platforms, AI chip design, memory chips and wafer foundry. It is worth noting that several of these companies are also Anthropic's computing power suppliers or important industry partners.

Previously, Anthropic sent a listing document to underwriters, stating that the company plans to issue about 867 million Class A ordinary shares, with a proposed issue price range of $70 to $80 per share. Calculated at the upper limit of the issue price, the company plans to raise about $69.3 billion through the public offering.

However, according to Reuters, Anthropic later sent an email to underwriters saying that the pricing information in the previous document was inaccurate and that the relevant documents would be updated. Therefore, the number of shares to be issued, the price range and the scale of fundraising still need to be further confirmed by the company.

Anthropic's listing valuation is facing adjustment pressure. Several months ago, the market discussed the possibility of Anthropic listing at a valuation of $2 trillion or even higher. But as AI concept stocks recently suffered a sharp sell-off, investors' expectations have clearly fallen back.

According to Bloomberg, some potential investors are discussing a valuation range of about $1.2 trillion to $1.3 trillion, and some are even considering pushing the valuation down to $1.1 trillion. Previously, the reasonable valuation expectations investors expressed to underwriting banks were about $1.4 trillion to $1.5 trillion.

Anthropic's shares may begin trading as early as before November 19 or 20, completing the listing before the U.S. Thanksgiving holiday on November 26.

Two frontier models launched on the eve of the IPO, focusing on improving coding, reasoning and visual capabilities

As preparations for the listing enter a critical stage, Anthropic is also accelerating product iteration in an attempt to further consolidate its competitiveness in the frontier AI model market.

On the 9th, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, focusing on improving programming, reasoning and visual capabilities. According to the introduction, the new models' output speed is up to 30% faster than the previous generation, while usage prices have been lowered, charging $8 per million input tokens and $40 per million output tokens.

Anthropic emphasized that the new models are not only faster, but can also complete work with fewer tokens on complex tasks, reducing the cost of individual tasks. In addition, the two models also offer an optional "fast mode," with output speed increased to up to 2.5 times the original, while input and output prices double.

The two models are based on the same underlying model but differ in safety protections and scope of availability. Among them, Fable 5.2 is open to a broader range of users, while Mythos 5.2 is mainly open to vetted cybersecurity and life sciences research institutions.

Anthropic said Fable 5.2 is better suited to handling complex, open-ended technical tasks and has strong capabilities in fields such as engineering, mathematics and robotics. The new models released this time have undergone rigorous safety and alignment testing, and new evaluation tools and post-release monitoring mechanisms have been developed.

Cases published by Anthropic show that Fable 5.2 can coordinate more than 1,000 sub-agents, work continuously for 8 days to autonomously develop an operating system, and complete a formal proof of Fermat's Last Theorem in 30 hours.

Fable 5.2 achieved excellent results in the FrontierCode 2.0 test, which measures complex programming capabilities. Source: Anthropic

It is worth noting that this is the first time Anthropic has released a frontier model since Amodei called for slowing the pace of AI research and development. Facing external concerns about the speed and safety of AI research and development, Dianne Penn, head of research product management at Anthropic, said: "What we called for is controlling the pace of research and development, not pausing it."

From a financial performance perspective, Anthropic's growth remains rapid, but computing power costs are also continuing to climb. The company's latest prospectus shows that revenue in the third quarter of 2026 reached $16.956 billion, up 47% quarter-on-quarter, lower than the 144% quarter-on-quarter growth rate in the second quarter; during the same period, the company had an operating loss of $567 million and a net loss of $23.291 billion.

In contrast, the total amount of Anthropic's signed computing power and infrastructure procurement commitments and lease agreements has reached $895.4 billion, of which about $447.5 billion is expected to occur before the end of 2030. Responsible editor: Zheng Jingxin Image editor: Jin Jie Proofreader: Zhang Yan The Paper tip-off: 021-962866 The Paper, unauthorized reproduction prohibited +1 Favorite I want to report #Anthropic #AI #ArtificialIntelligence View more