Anthropic has officially launched two new advanced AI models: Claude Fable 5 and Claude Mythos 5. This marks a significant expansion of Anthropic's powerful 'Mythos-class' AI capabilities, which were previously only accessible to select organizations through its Project Glasswing cybersecurity program. The introduction of Fable 5 aims to provide broader access to these enhanced features for general users and developers, while Mythos 5 continues to serve a more restricted group for specialized, high-risk applications.
Claude Fable 5 represents a substantial upgrade over previous general-release Claude models. It demonstrates superior performance across a wide range of demanding tasks, including software engineering, complex knowledge work, sophisticated scientific research, and extended analytical processes. While the prior Claude Mythos Preview model still holds top positions in specific niche benchmarks like computer use and multidisciplinary reasoning, Fable 5 exhibits broad improvements across most key performance indicators, setting a new standard for general-purpose AI models.
Enhanced Performance and Strategic Deployment
Fable 5 is engineered to exceed the performance of its predecessors in nearly all general applications. Its capabilities are designed to offer robust support for software development, scientific inquiry, and complex problem-solving. The integration of these advanced features into a widely accessible model signals Anthropic's commitment to democratizing cutting-edge AI technology, while maintaining a strategic approach to managing potentially sensitive applications.
The benchmark comparison chart highlights the advancements made with Fable 5 and Mythos 5, illustrating their performance against existing models. This data provides a quantitative measure of the leap in capabilities, particularly in areas crucial for enterprise and research applications.
Controlled Access for Specialized Applications
Claude Mythos 5, on the other hand, is positioned as a less restricted, more potent version of the Mythos-class AI. While it offers broader capabilities, its availability is intentionally limited. It is primarily accessible to Anthropic-approved users, including cybersecurity partners within Project Glasswing and select biology researchers. This tiered release strategy allows Anthropic to deploy its most powerful AI tools responsibly, ensuring that access is granted based on specific needs and risk assessments.
API documentation and model information screens like this one on Anthropic's website are crucial for developers and enterprise users seeking to integrate these advanced AI systems. The detailed specifications help in understanding the deployment options and performance characteristics of models like Fable 5 and Mythos 5.
Pricing, Access, and Rollout Strategy
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens. While this pricing is less than half that of the Claude Mythos Preview, it positions them as premium offerings in the current AI market. This pricing structure reflects the advanced capabilities and extensive development invested in these models.
Model | Input | Output | Total Cost | Source |
MiMo-V2.5 Flash | $0.10 | $0.30 | $0.40 | Xiaomi MiMo |
deepseek-v4-flash | $0.14 | $0.28 | $0.42 | DeepSeek |
deepseek-v4-pro | $0.435 | $0.87 | $1.305 | DeepSeek |
MiniMax-M3 | $0.30 | $1.20 | $1.50 | MiniMax |
Gemini 3.1 Flash-Lite | $0.25 | $1.50 | $1.75 | |
Qwen3.7-Plus | $0.40 | $1.60 | $2.00 | Alibaba Cloud |
MiMo-V2.5 | $0.40 | $2.00 | $2.40 | Xiaomi MiMo |
Grok 4.3 (low context) | $1.25 | $2.50 | $3.75 | xAI |
GLM-5 | $1.00 | $3.20 | $4.20 | Z.ai |
Kimi-K2.6 | $0.95 | $4.00 | $4.95 | Moonshot/Kimi |
GLM-5.1 | $1.40 | $4.40 | $5.80 | Z.ai |
Grok 4.3 (high context) | $2.50 | $5.00 | $7.50 | xAI |
Qwen3.7-Max | $2.50 | $7.50 | $10.00 | Alibaba Cloud |
Gemini 3.5 Flash | $1.50 | $9.00 | $10.50 | |
Gemini 3.1 Pro Preview (≤200K) | $2.00 | $12.00 | $14.00 | |
GPT-5.4 | $2.50 | $15.00 | $17.50 | OpenAI |
Gemini 3.1 Pro Preview (>200K) | $4.00 | $18.00 | $22.00 | |
Claude Opus 4.8 | $5.00 | $25.00 | $30.00 | Anthropic |
GPT-5.5 | $5.00 | $30.00 | $35.00 | OpenAI |
Claude Fable 5 / Claude Mythos 5 | $10.00 | $50.00 | $60.00 | Anthropic |
For developers, Fable 5 is accessible via the Claude API as claude-fable-5 and is included in consumption-based Enterprise plans. The rollout for subscription users is phased: Fable 5 is available at no extra cost on Pro, Max, Team, and Enterprise plans until June 22. Post this date, its use will require usage credits, with plans to reintegrate it as a standard feature as soon as feasible.
Differentiating Fable 5 and Mythos 5
The distinction between Fable 5 and Mythos 5 is not one of inherent capability size but rather of access control and embedded safeguards. Both models are built on the same underlying Mythos-class architecture. Fable 5 incorporates an additional layer of safety protocols designed to identify and reroute requests related to high-risk domains—including cybersecurity, biology, chemistry, and model distillation—to Claude Opus 4.8. Users are notified when such a redirection occurs.
Mythos 5 removes some of these restrictions, making it more suitable for vetted users operating in specialized fields. For general enterprise and developer tasks, Fable 5 is designed to perform comparably to Mythos 5. This dual approach allows Anthropic to deploy frontier AI models by balancing broad utility with the responsible management of high-risk functionalities.
Significant Advancements in Autonomous Coding
A primary application for enterprise adoption of Fable 5 is in software engineering. Anthropic reports that Fable 5 demonstrates enhanced autonomy and can operate for extended periods without direct human supervision, a critical feature for AI agents tasked with complex coding projects. On the SWE-bench Pro benchmark, which assesses a model's ability to tackle challenging software engineering tasks, Fable 5 and Mythos 5 achieved an impressive 80.3% success rate, significantly surpassing GPT-5.5's score of 58.6%.
Further demonstrating its prowess, Fable 5 achieved a 29.3% score on Cognition’s FrontierCode Diamond benchmark for high-quality, agentic coding. This performance far exceeds that of Claude Opus 4.8 (13.4%) and GPT-5.5 (5.7%). Early user testimonials highlight Fable 5's impact: Stripe reported that the model executed a codebase-wide migration on their 50-million-line Ruby codebase in a single day, a task that would typically take over two months for a human team. Similarly, Cursor noted Fable 5's ability to handle complex, long-horizon problems previously out of reach for earlier models.
Elevated Performance in Knowledge Work and Analysis
Beyond coding, Fable 5 shows marked improvements in knowledge work applications. On the GDPval-AA benchmark, Fable 5 and Mythos 5 scored 1932, surpassing Claude Opus 4.8 (1890) and GPT-5.5 (1769). The models also excel in document reasoning tasks, achieving 29.8% on the GDPpdf benchmark without tools, a notable improvement over Opus 4.8 (22.5%) and GPT-5.5 (24.9%).
These gains are particularly relevant for enterprises dealing with unstructured data, such as PDFs, reports, and contracts. Hex reported Fable 5 broke 90% on their analytics benchmark for complex, long-running tasks, a 10-point leap over Opus. Hebbia noted Fable 5 as the top performer on its Finance Benchmark for senior-level reasoning. Financial institutions like IMC and Optiver have reported strong performance in trading-analysis evaluations, attesting to the model's accuracy in factual lookup, conceptual reasoning, and root-cause analysis. Legal sector evaluations by Crosby Legal suggest Fable 5's redlining capabilities match or exceed current models, while Notion and Zapier highlight its ability to transform raw notes into structured project plans and enhance automation workflows.
Enhanced Vision and Interface Understanding Capabilities
Anthropic reports that Fable 5 also features its most advanced vision capabilities to date. The model can accurately extract numerical data from complex scientific figures and perform vision-based tasks, such as reconstructing application source code solely from screenshots. This enhanced visual understanding has significant implications for enterprise automation, enabling AI agents to interact more effectively with visual interfaces like dashboards, legacy applications, and image-heavy reports that lack direct API integrations.
In internal tests, Fable 5 demonstrated remarkable agentic skills by successfully playing the game Pokémon FireRed using a minimal vision-only harness, a feat that previous Claude models struggled with. This showcases the model's ability to perceive, strategize, and execute over extended periods. Furthermore, when integrated with persistent file-based memory in the game Slay the Spire, Fable 5 showed a threefold performance improvement compared to Opus 4.8, reaching the game's final act more frequently. These advancements suggest Fable 5's potential to leverage contextual information more effectively in multi-step enterprise workflows.
From Restricted Cybersecurity Model to General Enterprise AI
The launch follows Anthropic's introduction of Claude Mythos Preview via Project Glasswing in April 2025, a program designed for cybersecurity defenders and critical infrastructure providers. Mythos-class models were initially developed recognizing their potential for both advanced cyber defense and offensive capabilities, prompting concerns about misuse. The potential for AI-assisted vulnerability discovery led to discussions among U.S. officials, intelligence agencies, and financial regulators regarding its impact on cyber operations and financial stability.
While governments like South Korea have sought access to Mythos for national cyber defense, concerns about secure access have persisted. The development of Fable 5 represents Anthropic's strategy to commercialize Mythos-class models by distinguishing general enterprise value from the riskiest functionalities. Fable 5 is equipped with classifiers that detect and reroute requests in three key areas: cybersecurity (vulnerability discovery, reconnaissance, lateral movement), biology and chemistry (potential for misuse in dangerous research), and model distillation (preventing extraction of capabilities for competitor models). Requests flagged by these classifiers are handled by Claude Opus 4.8, with users being informed of the redirection.
Anthropic reports extensive internal and external red-teaming efforts, with over 1,000 hours of testing yielding no universal jailbreaks. While Fable 5 demonstrates a commitment to safety, Anthropic acknowledges that its cautious safeguards may occasionally trigger on legitimate, benign requests, particularly from security professionals and researchers. The company plans to refine these classifiers to minimize false positives over time.
Mythos 5: The Frontier for Specialized Sectors
Mythos 5 is designated for enterprises operating in sensitive sectors such as cybersecurity, critical infrastructure, and life sciences. Current users with Claude Mythos Preview access can upgrade, with plans for expanded access through a trusted user program in collaboration with the U.S. government. This model is critical for sectors where capabilities flagged by Fable 5's safeguards are core operational requirements, such as security teams needing to reproduce vulnerabilities or biology researchers exploring complex molecular designs.
Anthropic claims Mythos 5 exhibits the world's strongest cybersecurity capabilities, scoring 78.0% on ExploitBench, significantly higher than existing models. In biology, Mythos-class models have shown potential to accelerate drug design processes by up to tenfold, identifying binding sites and utilizing bioinformatics tools with minimal human assistance. These advancements suggest that Anthropic's high-end models are capable of performing complex scientific research tasks with increased autonomy.
New Data Retention Policy for Advanced Models
A notable policy change accompanying the Fable 5 and Mythos 5 launch is a mandatory 30-day data retention period for all traffic associated with these models and future high-capability successors. This policy applies across first-party and third-party platforms. Anthropic assures that this data will not be used for training new models or non-safety purposes, and has implemented privacy protections including logging human access and data deletion after 30 days. This requirement introduces a key consideration for enterprises, particularly those in regulated industries, regarding data governance and compliance.
Broader Enterprise Implications and Competitive Landscape
The introduction of Fable 5 signifies Anthropic's strategy to commercialize highly autonomous AI models by implementing tiered access and domain-specific restrictions. This approach could establish a new paradigm for releasing advanced AI systems, balancing broad market utility with the careful management of potent capabilities. For enterprises, Fable 5 promises enhanced capabilities in autonomous coding, knowledge work, and complex analytical tasks, potentially streamlining workflows and improving operational efficiency.
The differentiated access model for Fable 5 and Mythos 5 presents a strategic choice for various enterprise sectors. While Fable 5 offers general-purpose advancements, Mythos 5 provides restricted access to specialized, high-risk capabilities for vetted entities. This tiered approach aims to maximize the benefits of frontier AI while mitigating associated risks. The competitive landscape is also shaped by these advancements, as Anthropic asserts state-of-the-art performance across multiple benchmarks, challenging existing industry leaders and potentially accelerating the adoption of more powerful AI solutions in enterprise environments.
