Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, cutting the price of its most demanding AI work while claiming stronger results in coding, reasoning and visual tasks. The launch comes as the company pursues a stock-market debut with a reported valuation target centered on $1.5 trillion.
For developers paying by usage, the change is straightforward: both models cost $8 per million input tokens and $40 per million output tokens. That's 20% below the $10 and $50 rates for their 5.1 predecessors. Tokens are the pieces of text that models process and generate.
The company's announcement says the new versions generate output up to 30% faster. An optional fast mode can increase token-generation speed by up to 2.5 times, but costs twice the standard rate: $16 per million input tokens and $80 per million output tokens.
Lower prices, different access rules
Fable and Mythos use the same underlying model. The difference is their safeguards and who can use them. Fable has cybersecurity and biology restrictions intended to limit misuse; organizations seeking Mythos for cybersecurity or biomedical work must apply through verification programs.
Fable 5.2 is available across Anthropic's platforms and through Amazon Web Services, Google Cloud and Microsoft Azure. Enterprise customers can also access Mythos through Claude Security, which scans code for vulnerabilities.
For ordinary coding and knowledge tasks, the company recommends starting with Claude Opus 5.5. It describes the role of its premium models this way:
Fable and Mythos are best for the most ambitious, open-ended technical work.
Its published pricing puts Opus 5.5 at $4 per million input tokens and $20 per million output tokens. Fable's lower bill is therefore a reduction within its own premium tier, rather than the lowest price in the Claude lineup.
The performance gains also depend on how much effort users ask the model to spend. In Anthropic's FrontierCode 2.0 coding tests, Fable 5.2 scored 49.1% at its Xhigh setting, compared with 32.4% for Fable 5.1. At High effort, the new version roughly matched Opus 5.5's best score for about half the cost per task. Those are company-reported test results.
A product launch ahead of the share sale
As previously reported by Apex, Anthropic is targeting a listing before Thanksgiving on November 26, with fundraising expectations of $80 billion to $100 billion and a valuation range of $1.25 trillion to $1.75 trillion. The $1.5 trillion figure sits at that range's midpoint; it isn't a finalized IPO valuation.
Anthropic's registration announcement says it intends to trade on the Nasdaq Global Select Market under ANT. It also leaves the offering's central terms open:
The number of shares of Class A common stock to be offered and the price range for the proposed offering have not yet been determined.
The business behind that proposed listing is growing quickly and making large spending commitments. Apex previously reported that its amended prospectus showed $33.2 billion in revenue for the nine months ended September 30, alongside approximately $895.4 billion in committed spending, substantially all for computing and infrastructure.
The new release gives prospective customers more detail on what that computing can do. Anthropic says prerelease work included migrating a production service from Python to Rust and designing three working electronic devices. It also reports progress in robotics and scientific research, including two open problems involving black-hole physics.
There are limits to the upgrade. Anthropic says it deliberately constrained gains in certain capabilities that could accelerate AI development, including accelerator-kernel optimization and improvements to large-language-model architectures. In those areas, it puts the new models at roughly Opus 5.5's level.
Access comes with data-handling conditions, too. Anthropic is testing Enterprise Frontier Safeguards, which runs automated safety monitoring on prompts and responses stored within a customer's infrastructure. Eligible enterprise customers can continue using the models with zero data retention before implementing that monitoring. Other customers face 30-day retention for safety monitoring.