New Model Sets The Bar For AI Safety And Efficiency

Anthropic has launched Claude Fable 5, the latest public version of its Mythos model, expanding access to a system designed for software engineering, knowledge work and computer vision tasks. The company said high-risk requests involving areas such as cybersecurity, biology, chemistry and AI model distillation will be redirected to Claude Opus 4.8, which has been configured with additional safeguards.

Strategic Rollout And Broader Accessibility

Mythos was initially made available to a limited group of partners in April as Anthropic evaluated potential cybersecurity risks associated with the model. Access was expanded last week to hundreds of organisations across 15 countries, primarily those operating critical infrastructure. Claude Fable 5 is now available through Anthropic’s Claude API and usage-based Enterprise plans. Early access has also been included in selected subscription tiers ahead of a broader pricing rollout scheduled for June 23.

Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram

Advancing Safety And Industry Standards

Anthropic said the model underwent extensive safety testing before release, including bug bounty programmes and red-team exercises conducted by external organisations. According to the company, more than 1,000 hours of testing did not identify any universal jailbreak vulnerabilities.

A mandatory 30-day data retention policy will apply to all traffic processed by the model, including accounts that previously operated under zero-retention agreements. Anthropic said the measure is intended to improve monitoring and protection against emerging security threats.

Outstanding Performance And Competitive Pricing

Independent evaluations, including testing by analytics company Hex, reported strong performance in complex reasoning and analytical tasks. Companies, including Base44 and Genspark, highlighted improvements in tool use and interface design capabilities. Pricing has been set at $10 per million input tokens and $50 per million output tokens, compared with lower rates for previous models. Some enterprise customers, including Rakuten, said the model’s ability to verify aspects of its own output could help improve efficiency in tasks that require higher levels of accuracy.

Implications For The AI Market

The release comes as Anthropic prepares for a potential public market debut, and competition among leading AI developers continues to intensify. Alongside performance improvements, the company has placed significant emphasis on model safety, reflecting broader industry concerns around misuse, jailbreak attempts and the risks associated with increasingly capable AI systems.