Anthropic Claude Fable 5 and Mythos 5 Overview

Model Performance and Benchmarks

Claude Fable 5 is a Mythos-class model that Anthropic claims exceeds the capabilities of its previous general-purpose models. Performance gains are most notable in coding tasks, with the model scoring over double the results of Claude Opus 4.8 on the Frontier Code benchmark. It also shows a 30% improvement in legal reasoning tasks compared to Opus 4.8. However, performance on tool use and computer-use benchmarks remains comparable to existing models, suggesting that the intelligence gains are not uniform across all domains.

Safety, Data, and Access Restrictions

Anthropic has implemented strict safety classifiers that automatically switch the model to Claude Opus 4.8 if a query triggers concerns regarding cybersecurity, biology, or attempts to extract the model's chain-of-thought process. These classifiers monitor the entire conversation history rather than just the latest prompt. Additionally, the company has introduced a mandatory 30-day data retention policy for all traffic on Mythos-class models, even for third-party surfaces, to facilitate safety monitoring and prevent jailbreaks. Access to Fable 5 is currently included in Pro and Max subscription plans until June 22, after which users will likely transition to pay-per-token API pricing.

Pricing and Economic Model

While Fable 5 is priced at double the cost of Claude Opus 4.8, it remains significantly cheaper than the initial Claude Mythos preview. The pricing is set at $10 per million tokens for input and $50 per million tokens for output. The shift toward token-based billing for premium models suggests that Anthropic is moving away from flat-rate subscription models for its most capable inference tiers.