Anthropic Unveils Upgraded Fable 5.2 and Mythos 5.2 Models as Public Listing Nears

New models arrive ahead of a major listing

Anthropic has introduced upgraded versions of two of its flagship artificial intelligence systems, Fable 5.2 and Mythos 5.2, just weeks before it plans to go public. The San Francisco-based company says both models run considerably faster than their predecessors, with output generated up to 30 percent quicker, and they are priced at roughly a fifth less, a shift that could matter for businesses weighing AI costs.

Two versions, one underlying model

Although they are marketed separately, the two offerings are built on the same core model. Fable 5.2 includes protective filters that block the majority of cybersecurity and biology-related requests, allowing Anthropic to make it broadly accessible. Mythos 5.2 operates with far fewer such limits and is restricted to vetted organisations engaged in cybersecurity or biomedical research. The company has said Fable 5.2 is available to paid Claude subscribers.

Performance claims

Anthropic’s own benchmark figures place Fable 5.2 ahead of its earlier release and several rival systems in coding evaluations. On the FrontierSWE v2 agentic coding test, it recorded 69.9 percent, compared with 56.3 percent for Fable 5.1, 65.5 percent for OpenAI’s GPT-6 Astra and 61.2 percent for Google’s Gemini 4 Carbon. In a business workflow test, AutomationBench, Google’s Gemini 4 Carbon came first at 53.9 percent, with Fable 5.2 close behind at 51.4 percent. Mythos 5.2 posted 74.6 percent on the Terminal-Bench-Science 0.1 research benchmark, ahead of the two competing models. The company also described an internal experiment in which Fable 5.2, coordinating more than 1,000 subagents over eight days, produced a working prototype operating system. These results come from the company and have not been independently verified.

Earlier setbacks

The original Fable and Mythos models had a turbulent debut in June. Days after launch, the US Commerce Department cited national security concerns and directed Anthropic to block access for foreign nationals, forcing both models offline. Those restrictions were lifted several weeks later.

Financial picture and IPO plans

Anthropic is aiming to list its shares the week of November 16 and could raise as much as $100 billion, which would be the largest initial public offering on record. Its valuation has been projected above $1 trillion, though expectations have come down from an earlier estimate of $2 trillion. Revenue climbed to $17 billion in the third quarter from $1.4 billion a year earlier, according to a recent securities filing. Growth nonetheless slowed compared with earlier quarters, as competition from OpenAI and lower-cost models, many from China, intensified. The company also reported an operating loss of $567 million for the quarter, driven by higher spending on short-term computing capacity.

Background

The listing is being watched as a barometer for the durability of the AI investment wave that has supported the US economy. Investor sentiment has been dampened by a broader market pullback and concern over the very large sums technology firms are putting into AI. Political developments add another layer of uncertainty: Democrats regained control of Congress in the midterm elections, prompting worries about possible tighter rules for the technology sector, including data centre construction. Ballot measures limiting such construction could also affect Anthropic and rivals such as OpenAI, Meta, Microsoft and Google.

What next

Attention now turns to whether the company proceeds with its planned November listing and how the market values it. Investors will also be looking to see whether the performance gains claimed for the new models translate into sustained revenue growth in a crowded and price-sensitive market.