Claude Fable 5 and Claude Mythos 5: A Revolutionary Leap in AI Capabilities and Safety
Anthropic, a leading AI research company, has unveiled the latest advancements in its Claude series with the release of Claude Fable 5 and Claude Mythos 5. These models represent a significant leap forward in AI capabilities, pushing the boundaries of what is possible in various domains, while also addressing critical safety concerns.
Unparalleled Capabilities
Fable 5 boasts capabilities that surpass those of any previous Claude model. It excels in software engineering, knowledge work, vision, scientific research, and more. In software engineering, Fable 5 can compress months of work into days, as demonstrated by Stripe's experience with codebase-wide migrations. It also demonstrates superior token efficiency, scoring highest on the FrontierCode evaluation.
In knowledge work, Fable 5 shines with strong performance on complex analytical tasks. It has the highest score on Hebbia's Finance Benchmark for senior-level reasoning and aced trading-analysis evaluations from IMC. The model's reasoning capabilities are on par with senior research scientists, making it a valuable asset for complex problem-solving.
In the realm of vision, Fable 5 is a state-of-the-art model. It can extract precise numbers from detailed scientific figures and perform complex vision-based tasks, such as rebuilding a web app's source code from screenshots. Its ability to play games like Pokémon FireRed with minimal scaffolding showcases its versatility.
Fable 5's memory and long-context capabilities are also remarkable. It stays focused across millions of tokens and improves its outputs using its own notes. When playing the deck-building game Slay the Spire, Fable 5's performance was significantly enhanced by persistent file-based memory.
Drug Design and Scientific Research
Mythos 5, a variant of Fable 5, has accelerated drug design processes by around ten times. It can match or beat skilled human operators in tasks like protein design and bioinformatics. The model's ability to execute tasks typically performed by scientists, such as choosing binding sites and running protein design tools, is remarkable.
In molecular biology, Mythos 5 consistently produces novel and compelling scientific hypotheses. It has advanced several hypotheses to experimental evaluation, and one hypothesis was corroborated in a study from a lab working on the same problem. The model's research in genomics is equally impressive, with Mythos 5 conducting novel genomics research in over a week of autonomous work.
Safety and Alignment
Anthropic recognizes the risks associated with releasing such powerful models. They have implemented robust safety measures, including classifiers that detect potential misuse and prevent the main model from responding. These classifiers cover cybersecurity, biology and chemistry, and distillation, ensuring that Fable 5 falls back to Claude Opus 4.8 when necessary.
The company has also introduced a new data retention policy for business customer data, requiring 30-day retention for Mythos-class models. This data will be used for safety-related purposes and to defend against complex attacks.
Availability and Pricing
Claude Fable 5 is available everywhere, while Claude Mythos 5 is initially restricted to Glasswing partners and select biology researchers. Both models are offered at $10 per million input tokens and $50 per million output tokens.
Anthropic aims to expand access to Claude Mythos 5 through a trusted access program, allowing cybersecurity organizations to apply systematically. They also plan to open a trusted access program for biology, providing access to Fable 5 with biology and chemistry safeguards removed.
Conclusion
The release of Claude Fable 5 and Claude Mythos 5 marks a significant milestone in AI development. These models demonstrate the incredible progress in AI capabilities while also addressing safety concerns. As Anthropic continues to refine its safeguards and expand access, the potential for these models to revolutionize various industries becomes increasingly evident.
