Anthropic launches Claude Fable 5 and Mythos 5, the world's most capable AI models.

Anthropic has just raised the level of artificial intelligence with the launch of Claude Fable 5, considered the most capable model ever available to the public. It surpasses all competitors in software engineering benchmarks, knowledge work, computer vision and scientific research, with increasing advantage in more complex tasks. Simultaneously, the company released Claude Mythos 5, a version with reduced safeguards aimed exclusively at cyberdefenders and critical infrastructure providers, in partnership with the US government.

What makes Claude Fable 5 of Anthropic so powerful?

Fable 5 reaches the state of the art in virtually all standardized AI capacity tests. In software engineering, for example, it demonstrates deep understanding of legacy code and ability to generate optimized solutions for complex problems. In the areas of vision and scientific research, the model interprets graphs, diagrams and technical articles with precision comparable to that of human experts. The greater and more ambiguous the task, the greater the advantage over previous models, including Claude Opus 4.8.

However, Anthropic recognizes that such a capable model requires strict controls. "Without safeguards, Fable 5's capabilities in areas such as cybersecurity could be misused to cause serious damage," says the company in its official statement. Therefore, consultations on sensitive topics are automatically redirected to Claude Opus 4.8, which acts as a more conservative response model.

Why do Claude Fable 5's safeguards generate false positives?

To launch the model quickly and with acceptable minimum security, Anthropic adopted a conservative approach: protections fire in less than 5% of sessions, but occasionally block perfectly harmless requests. The company promises to reduce these false positives in the coming weeks as new models are launched. This strategy recalls the challenges faced by other companies in balancing security and usability — an increasingly urgent issue, as the recent case of the self-replicant AI worm that threatens cyber security shows.

What is Claude Mythos 5 and who can use it?

Claude Mythos 5 shares the same basis as Fable 5, but with safeguards relieved in specific areas, mainly related to cybersecurity. He is the model with the greatest cyberdefense capability ever created, according to Anthropic. Initially, it will be implemented via Project Glasswing, in collaboration with the US government, as an update of Claude Mythos Preview. In the future, access will be extended through a trust program. This shows how industry is migrating to specialized and controlled models, in a trend that AI companies expand to rival markets also exploit.

How does Claude Fable 5 impact Brazilian developers and professionals?

For the Brazilian market, Fable 5 represents a leap in productivity in areas such as software development, data analysis and process automation. Professionals who deal with complex code, technical documentation and scientific research may obtain more accurate results. Mythos 5, while restricted to authorized entities, signals that cybersecurity will gain high performance AI tools — something essential for companies dealing with sensitive data. Meanwhile, Anthropic continues to refine its safeguards to reduce false positives that still hinder daily tasks.

Frequently Asked Questions about Anthropic Claude Fable 5

Is Claude Fable 5 available to all users?

Yes, Fable 5 is Anthropic's standard model from now on, accessible via API and company products, with the described safeguards.

Can Mythos 5 be used by any security company?

Nope. Initially, only Project Glasswing participants and US government partners have access. Anthropic plans to expand to a reliable access program in the coming months.

What are the main benchmarks in which Fable 5 stands out?

The model leads software engineering, knowledge work, computer vision and scientific research, overcoming all competitors in metrics such as SWE-bench, MMLU and others.

Fonte: Anúncio oficial da Anthropic