Anthropic has launched Fable 5 and Mythos 5, two new AI models, expanding their capabilities and access for various applications.
On June 9, Anthropic unveiled two new models in its Claude family: Fable 5, designed for general use, and Claude Mythos 5, a more restricted version intended for private applications.
The primary distinction between these models is their protective classifiers. Fable 5, available through a standard subscription, is programmed to monitor sensitive queries, particularly in fields like cybersecurity and biology, redirecting potentially harmful queries to Claude Opus 4.8. The company noted that these protective measures trigger in less than 5% of sessions, sometimes even responding to benign inquiries.
“To release the model safely and quickly, we conservatively tuned these protective measures,” Anthropic stated.
Both models are priced identically at $10 for 1 million input tokens and $50 for 1 million output tokens, which is significantly less than the earlier Claude Mythos Preview.
Access and Capabilities of Mythos 5
Claude Mythos 5 is currently accessible to a select group of trusted participants, including members of Project Glasswing, and serves as a successor to the Claude Mythos Preview. The company plans to broaden access for cybersecurity and biomedical tasks by the end of 2026.
While maintaining some restrictions, Mythos 5 is touted as the most advanced model for cybersecurity, excelling in identifying software vulnerabilities. It also demonstrates a significant improvement in drug development processes, achieving a tenfold speed increase in certain phases.
“Mythos 5 utilizes protein design and bioinformatics tools without human involvement, matching or often exceeding the capabilities of experienced human operators,” the company noted.
For biomedical researchers, Anthropic is developing a pathway to access Fable 5 without restrictions in biology and chemistry, although cybersecurity censorship will remain.
Public Launch of Fable 5
Anthropic claims that Fable 5 is the strongest public model, particularly adept at handling long and complex tasks such as programming, document analysis, and scientific research. However, the company has withheld specific technical details, including model architecture and parameter counts.
Early testing by Stripe highlighted Fable 5's efficiency, compressing what would have taken months of development into mere days. For instance, it successfully migrated a 50-million-line Ruby codebase in just one day.
Fable 5 also demonstrated superior token efficiency in the FrontierCode assessment, outperforming other AI solutions, and achieved the highest score in logical reasoning analysis during the Hebbia Finance Benchmark test.
Its advanced features include enhanced memory capabilities, allowing it to maintain focus over extensive tasks and utilize its own notes effectively. Notably, Fable 5 can extract precise data from complex scientific diagrams and perform advanced tasks with minimal assistance.
In early June, Anthropic expressed concerns about AI self-improvement, revealing that over 80% of the code for its current products was generated by Claude.
