Claude Fable 5: Anthropic’s Most Powerful AI for Everyone

Anthropic has unveiled two new artificial intelligence models, Claude Fable 5 and Claude Mythos 5, marking a significant expansion of its advanced “Mythos-class” capabilities. Previously accessible only to select participants in its Project Glasswing cybersecurity initiative, these powerful models are now being made available more broadly, signaling a new phase in Anthropic’s product strategy.

Claude Fable 5, designed for general users and developers, reportedly surpasses all previous generally available Claude models. It demonstrates enhanced performance across critical areas such as software engineering, knowledge work, vision processing, scientific research, and extended task execution. While Fable 5 excels across most benchmarks, the earlier Claude Mythos Preview version still holds a slight edge in computer use and multidisciplinary reasoning.

Claude Mythos 5, on the other hand, represents a more unrestrained version of the Mythos-class architecture, though its availability is intentionally limited. It is an evolution of the earlier Mythos Preview model, with certain safety protocols reduced. Access to Mythos 5 is officially restricted to Anthropic-approved users, including cybersecurity partners involved in Project Glasswing and select biological researchers.

The fundamental distinction lies in how Fable 5 incorporates the advanced Mythos-class capabilities within a more robust safety framework. Anthropic states that requests related to high-risk domains—such as cybersecurity, biology, chemistry, and model distillation—are automatically redirected to Claude Opus 4.8, Anthropic’s previous flagship model. Users are informed when such a redirection occurs. Mythos 5 does not implement this fallback mechanism.

Anthropic reports that over 95% of Fable 5 sessions are handled entirely by the model, without resorting to fallback responses. Extensive internal and external red-teaming efforts, spanning over 1,000 hours, have reportedly found no “universal jailbreaks” that could bypass its safeguards.

Fable 5 is now accessible to the public via Anthropic’s website, applications, and API. However, Mythos 5 will initially only be available to existing users of the Claude Mythos Preview.

Pricing, Access, and a Phased Rollout

Anthropic has set the pricing for both Fable 5 and Mythos 5 at $10 per million input tokens and $50 per million output tokens. While this represents a reduction of more than 50% compared to the Claude Mythos Preview, it remains at the higher end of global AI model pricing, as indicated by benchmark comparisons.

VentureBeat Frontier AI Model API Pricing Snapshot

For developers integrating via API, Fable 5 is available as `claude-fable-5` and is fully accessible today. It is also included in consumption-based Enterprise plans.

The rollout for subscription users is more nuanced. Fable 5 will be part of Pro, Max, Team, and seat-based Enterprise plans at no additional cost until June 22. Following this period, its use will require usage credits, with Anthropic aiming to reinstate it as a standard feature on subscription plans as soon as feasible.

Distinguishing Fable 5 and Mythos 5

Anthropic positions Fable 5 and Mythos 5 not as distinct models in terms of capability size, but rather as variants sharing the same core architecture, differentiated by access controls and embedded safety features. Fable 5 incorporates a layer of safeguards designed to detect and manage high-risk queries, including those in cybersecurity, biology, and chemistry, as well as model distillation attempts. Such queries are rerouted to Claude Opus 4.8, with user notification.

Mythos 5, conversely, relaxes some of these restrictions for approved users operating within specific domains. For most standard enterprise and developer tasks, Anthropic asserts that Fable 5 delivers performance comparable to Mythos 5. This approach allows Anthropic to introduce advanced models with potentially dual-use capabilities by segmenting access based on user vetting and defined risk profiles, while retaining the most potent capabilities for specifically authorized applications.

Significant Advances in Autonomous Coding

For businesses, the immediate impact of Fable 5 is most pronounced in software engineering. Anthropic highlights its enhanced ability to operate autonomously and for extended periods, a crucial feature for AI agents tasked with complex development projects. Benchmarks indicate substantial improvements:

• On the SWE-bench Pro, which assesses complex software engineering task completion, Fable 5 and Mythos 5 achieved an 80.3% success rate, significantly outperforming OpenAI’s GPT-5.5 (58.6%).
• On Cognition’s FrontierCode Diamond benchmark for agentic coding, the new models scored 29.3%, compared to 13.4% for Claude Opus 4.8 and 5.7% for GPT-5.5.

Early adopters have reported transformative results. Stripe, for instance, utilized Fable 5 for a codebase-wide migration across a 50-million-line Ruby codebase, completing in one day a task that would typically require a team over two months. Other users, including Cursor, Replit, and Figma, have noted Fable 5’s superior performance on long-horizon development tasks, agentic coding, and prototyping, enabling faster application development with fewer resources.

This advancement signifies Anthropic’s strategic push towards AI coding solutions capable of managing larger work units—from codebase migrations and app prototyping to pull request reviews and automated testing—moving beyond simple code completion.

Enhancements in Knowledge Work, Finance, Legal, and Operations

Anthropic is also positioning Fable 5 as a superior tool for enterprise knowledge work. Performance metrics on benchmarks like GDPval-AA show Fable 5 and Mythos 5 scoring 1932, surpassing Claude Opus 4.8 (1890) and GPT-5.5 (1769). In visual document reasoning (GDPpdf), the new models achieved 29.8%, a notable increase from Opus 4.8 (22.5%) and GPT-5.5 (24.9%).

These improvements are critical for businesses dealing with diverse document formats like PDFs, spreadsheets, and reports. Fable 5 demonstrates enhanced capabilities in document interpretation, data analysis, and complex problem-solving.

• Hex reported Fable 5 exceeding 90% on its core analytics benchmark for complex, long-running analytical tasks, a significant leap from Opus 4.8.
• Hebbia found Fable 5 to be the top performer on its Finance Benchmark, showing double-digit gains in document reasoning and chart interpretation.

In finance, Fable 5 is demonstrating potential for high-stakes analytical workflows, with IMC noting its strong performance in trading analysis and root-cause analysis, and Optiver highlighting its consistency and improvement over Opus 4.8. Legal and operations teams are also expected to benefit, with Notion and Zapier reporting enhanced efficiency in project planning and automation tasks, respectively.

Advancements in Vision and Interface Understanding

Anthropic claims Fable 5 represents its most advanced vision model to date. It can accurately extract data from complex scientific figures and perform tasks such as reconstructing web application code solely from screenshots. This capability has direct implications for automating business processes that rely on visual interfaces not easily accessible via APIs.

Furthermore, Fable 5 exhibits enhanced agentic capabilities, demonstrated by its ability to independently play and beat Pokémon FireRed using a vision-only interface. This highlights its capacity for environmental perception, decision-making, and long-horizon execution. Internal tests also showed Fable 5 significantly benefiting from persistent memory in games like Slay the Spire, suggesting improved performance in multi-step, context-aware enterprise tasks.

From Restricted Cybersecurity Tool to General Enterprise AI

The launch follows Anthropic’s April 2025 introduction of Claude Mythos Preview via Project Glasswing, a program tailored for cybersecurity professionals. Mythos-class models were initially developed with capabilities that raised concerns about potential misuse in discovering and exploiting software vulnerabilities.

The dual-use nature of these advanced AI capabilities has drawn attention from government bodies and regulatory agencies worldwide, spurring discussions on both cyber defense strategies and the acceleration of security patching. Anthropic faces the ongoing challenge of balancing the release of powerful AI advancements with robust safety protocols and responsible deployment.

With Fable 5, Anthropic aims to balance the broad enterprise utility of Mythos-class models against the inherent risks. The model is engineered to perform tasks across software engineering, research, and analysis while employing classifiers to identify and manage high-risk queries related to:

• Cybersecurity: Detecting vulnerabilities, performing reconnaissance, and simulating agentic hacking.
• Biology and Chemistry: Reasoning through complex scientific processes with potential dual-use applications.
• Model Distillation: Preventing the extraction of model capabilities to train less-safeguarded AI systems.

When Fable 5’s classifiers detect a high-risk query, the request is handled by Claude Opus 4.8. Anthropic reports that this system has undergone rigorous internal and external red-teaming, with no universal jailbreaks identified. However, the company acknowledges that these safeguards may occasionally flag benign requests, potentially impacting advanced users.

Mythos 5 and the Frontier of Restricted Access

Mythos 5 is specifically designed for enterprises in sectors like security, critical infrastructure, and life sciences, where the unrestricted capabilities are core to operational needs. Anthropic is expanding access through a trusted program in collaboration with the U.S. government.

In cybersecurity benchmarks, Mythos 5 demonstrates leading performance, achieving 78.0% on ExploitBench and 83.8% on CyberGym. Similarly, in biology, Anthropic’s internal experts have used Mythos 5 to accelerate drug discovery workflows, with promising results in identifying potential drug candidates and generating novel scientific hypotheses.

New Data Retention Policy

Anthropic has also implemented a new data retention policy requiring 30-day retention for all traffic associated with Fable 5, Mythos 5, and future high-capability models. This data will not be used for training new models or for non-safety purposes, and enhanced privacy protections, including logging human access and data deletion after 30 days, have been added. This policy may become a key consideration for enterprises, particularly those in regulated industries, balancing advanced AI capabilities with data control requirements.

Enterprise Implications

The launch of Fable 5 signifies Anthropic’s strategy to commercialize more autonomous AI models by offering tiered access and domain-specific restrictions. This approach could serve as a blueprint for deploying increasingly powerful AI systems across various industries.

For software development, Fable 5 promises to enable larger-scale autonomous tasks like code migrations and full application builds. In knowledge work, it aims to enhance efficiency in finance, legal, and operations by reducing the number of interactions required and improving clarity in ambiguous situations.

For security and life sciences sectors, the distinction between Fable 5 and Mythos 5 is crucial. While Fable 5 offers advanced general capabilities, Mythos 5 provides the unrestricted access needed for specialized tasks within these domains, available only to vetted users.

This release intensifies competition within the AI industry, with Anthropic claiming state-of-the-art performance across multiple benchmarks. The key innovation lies in its potential to enable the market release of more powerful AI models while maintaining control over the most sensitive capabilities. The ultimate success of Fable 5 will depend on its real-world reliability, cost-effectiveness, and the management of its data retention policies and potential governance challenges.

Business Style Takeaway: Anthropic’s dual release of Fable 5 and Mythos 5 establishes a new paradigm for deploying advanced AI, balancing cutting-edge capabilities with controlled access for high-risk applications. This strategy is crucial for enterprises seeking to leverage powerful AI for complex tasks like autonomous coding and advanced research, while navigating crucial considerations of data security and responsible AI deployment.

Learn more at : venturebeat.com