Anthropic has launched two new frontier AI models simultaneously – Claude Fable 5 and Claude Mythos 5. While most headlines focused on Fable 5, the more significant development is that Mythos 5 has now officially moved beyond the preview stage. The AI firm is positioning Fable 5 as the public-facing version of its new ‘Mythos-class’ architecture, while Claude Mythos 5 becomes the less restricted version available through a controlled-access program. Both models share the same underlying foundation and pricing – $10 per million input tokens and $50 per million output tokens – but differ in how Anthropic governs access to their most powerful capabilities.
The release follows months of anticipation surrounding Claude Mythos Preview, which Anthropic unveiled in April and immediately restricted after internal evaluations showed unprecedented cybersecurity capabilities. According to Anthropic, the model identified thousands of software vulnerabilities during testing and demonstrated a level of autonomous vulnerability discovery that prompted the company to limit access to around 200 organizations, including government agencies and critical infrastructure operators through its Project Glasswing initiative.
Meanwhile, Anthropic claims both Claude Mythos 5 and Claude Fable 5 deliver state-of-the-art performance across coding, reasoning, tool use, and domain-specific professional tasks. On SWE-Bench Pro, one of the industry’s most respected software engineering evaluations, the models scored 80.3%, outperforming Claude Mythos Preview (77.8%), Claude Opus 4.8 (69.2%), GPT-5.5 (58.6%), and Gemini 3.1 Pro (54.2%). The gains were even more prominent on the FrontierCode Diamond benchmark, where Mythos 5/Fable 5 achieved 29.3%, more than double Opus 4.8’s 13.4% and over five times GPT-5.5’s 5.7%.
In practical AI agent workflows, the models posted 17.4% on AutomationBench and 85.0% on OSWorld-Verified, placing them at or near the top of every major agentic benchmark shown in the company’s comparison. The strongest gains appeared in frontier reasoning and specialized professional domains. On Humanity’s Last Exam, a benchmark designed to remain difficult even for frontier models, Mythos 5/Fable 5 scored 59.0% without tools and 64.5% with tools, ahead of Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Anthropic also highlighted substantial improvements in biology, cybersecurity, and health-related evaluations, with scores of 46.1% on BioMysteryBench (hard), 78.0% on ExploitBench, and 66.0% on HealthBench Professional.
Claude Mythos 5 represents Anthropic’s most capable model ever released. Unlike Fable 5, it retains broader access to the advanced cyber reasoning abilities that made the preview system controversial. Anthropic is not making Mythos 5 generally available. Instead, access is being granted to selected cybersecurity organizations, infrastructure operators, and strategic partners under Project Glasswing. The program has already expanded beyond the United States and now includes organizations across more than 15 countries.
The launch also comes at a crucial time for Anthropic as the company prepares for an IPO and faces growing competition from other frontier AI labs. Recently valued at around $965 billion, the company is trying to commercialize its most powerful technology while also positioning itself as a leader in AI safety and controlled deployment. Rather than releasing a single model to everyone, the company has created a two-tier system – Claude Fable 5 for general public use and Claude Mythos 5 for approved organizations that need access to its highest-level capabilities.
The Tech Portal is published by Blue Box Media Private Limited. Our investors have no influence over our reporting. Read our full Ownership and Funding Disclosure →
Ashutosh is a Senior Writer at The Tech Portal, largely reporting on new tech, and intersection of technology and business. Ashutosh’s career spans across nearly a decade of technology writing across multiple platforms and languages.
