Anthropic Releases Claude Fable 5.2 and Mythos 5.2 With High Autonomous Capabilities

Anthropic officially released anthropic's claude Fable 5.2 and Mythos 5.2 on Monday, delivering advanced artificial intelligence models designed to turn brief instructions into functional software, physical object designs, and complex research solutions. The newly unveiled systems are capable of operating autonomously for days at a time, allowing users to build intricate applications simply by asking a chatbot.

AI content creator Matthew Berman tested the new technology extensively, describing Fable 5.2 as the best AI model he has used. His ambitious projects included fully playable video games and a customized browser-based version of Microsoft Excel equipped with formulas and pivot tables. However, the advanced capabilities come with substantial API costs accrued while the system autonomously writes code, tests functions, and resolves its own errors over extended periods.

Advanced AI models like Fable 5.2 can operate autonomously for days, creating complex software at significant API costs.

For instance, building a single game-inspired level took nearly 44 hours of runtime and processed over 8.6 billion tokens, resulting in an estimated developer API cost of $14,710. An Excel clone generated roughly $11,700 in expenses after cache discounts, while a Fall Guys-style game consumed more than 9 billion tokens over 71 hours, equating to about $13,000. Despite these high expenditures, testers noted that the projects still featured occasional bugs and awkward animations requiring further refinement.

For users with tighter budgets, the model can also handle smaller tasks economically. A simple orange cylinder was built for less than $40, and a water-balloon physics simulation cost $91. Anthropic noted that the standard developer API is priced at $8 per million input tokens and $40 per million output tokens, reflecting a price reduction compared to previous generations.

Understanding the New Models

The underlying architecture is shared between Fable and Mythos, though Fable includes tighter safeguards for biological and cybersecurity requests, while Mythos is restricted to verified organizations working in those specific fields. Anthropic reported that Fable 5.2 achieves a 49.1% success rate on its internal FrontierCode 2.0 evaluation, surpassing the previous 32.4% score.

Why it matters

The introduction of these autonomous systems changes how developers approach complex software creation and research tasks by handling multi-day coding processes independently. This shift primarily affects developers, enterprises, and technical creators who require heavy computational output, while standard users must navigate significantly higher usage costs for large-scale projects.