Boasting state-of-the-art performance across software engineering, vision, and life sciences, the new Mythos-class models introduce powerful capabilities alongside rigorous, dual-tiered safety protocols.
- A Dual-Release Strategy: Claude Fable 5 launches for general use with unprecedented performance across benchmarks, while Claude Mythos 5 brings unrestricted, top-tier capabilities to a trusted group of cyberdefenders and biological researchers.
- Innovative Safety Safeguards: To mitigate the risks of dual-use technology, Fable 5 employs aggressive new classifiers for cybersecurity, biology, and distillation, automatically routing high-risk queries to the highly capable Claude Opus 4.8.
- Accessible Power: Priced at less than half the cost of previous preview models, Fable 5 and Mythos 5 aim to democratize frontier AI, featuring a phased rollout plan to manage expected high demand.
Today marks a pivotal moment in the rapid evolution of artificial intelligence. In a joint launch, the highly anticipated Claude Fable 5—a Mythos-class model deemed safe for general use—and its unrestricted counterpart, Claude Mythos 5, have officially entered the market. As reporters covering the bleeding edge of AI development, it is clear that these models represent a massive leap forward in autonomous capability, but perhaps more importantly, a foundational shift in how tech companies are handling the deployment of high-risk, high-reward systems.
Fable 5 is not just an incremental update. According to extensive benchmark testing, it is the new state-of-the-art across nearly all metrics of AI capability. The model exhibits exceptional proficiency in software engineering, knowledge work, computer vision, and scientific research. Most notably, Fable 5’s lead over its predecessors widens significantly as tasks become longer and more complex, demonstrating a capacity for sustained, autonomous work that was previously unimaginable.
Groundbreaking Capabilities Across Industries
The capabilities of Fable 5 and Mythos 5 have already been pressure-tested by industry leaders, yielding results that could reshape entire sectors.
- Software Engineering: The model is a powerhouse for coding. During early testing, Stripe reported that Fable 5 compressed months of complex engineering work into mere days, successfully executing a massive codebase-wide migration on a 50-million-line Ruby project that would have taken a human team over two months. Furthermore, it currently holds the highest score among frontier models on Cognition’s FrontierCode evaluation.
- Advanced Knowledge Work: On complex analytical tasks, Fable 5 reigns supreme. It secured the top score on Hebbia’s Finance Benchmark for senior-level reasoning and aced trading-analysis evaluations for IMC, excelling in root-cause and expected-value analysis.
- Vision and Memory: Fable 5 sets a new standard for vision tasks, capable of rebuilding web app source code entirely from screenshots and navigating visually complex environments like Pokémon FireRed without extensive scaffolding. In long-context memory tests, such as playing the deck-building game Slay the Spire, persistent file-based memory allowed Fable 5 to reach the final act three times more often than older models.
- Life Sciences and Genomics: Mythos 5 is already proving to be a revolutionary tool for molecular biology. Internal experts used the model to accelerate drug design by a factor of ten, with the AI operating entirely autonomously to choose binding sites and run protein design tools. The model also generated a novel hypothesis regarding an E. coli protein mechanism—which was later independently corroborated by a human lab—and successfully trained a custom machine learning model for genomics research that outperformed a model recently published in Science.
The Safety-First Approach: A New Paradigm
Releasing a model with the capacity to autonomously engineer software and design viral vectors comes with undeniable risks. The potential for malicious uplift—where bad actors use AI to accelerate cyberattacks or develop bioweapons—is a serious concern.
To address this, Fable 5 has been launched with heavily tuned, conservative safeguards. Central to this safety apparatus is a new suite of classifiers. If these secondary AI systems detect a high-risk query related to cybersecurity, biology/chemistry, or AI distillation, Fable 5 is blocked from answering.
Rather than issuing a frustrating, flat refusal, the system seamlessly falls back to Claude Opus 4.8. This ensures users still receive a highly capable response. Early data suggests these classifiers trigger in less than 5% of sessions, meaning the vast majority of users will experience Fable 5’s full, unhindered power.
These safeguards are incredibly robust. External red-teaming, including a bug bounty program with over 1,000 hours of testing, failed to find any universal jailbreaks. Even against 30 different public jailbreak techniques, Fable 5 complied with zero harmful requests relating to cyberattack planning or exploit development.
Privacy, Data Retention, and Trusted Access
Alongside the model launch comes a significant shift in data policy. For Fable 5, Mythos 5, and future models of this caliber, a mandatory 30-day data retention policy will be enforced for all business traffic. While strictly ring-fenced from being used for future model training, this logged data will be utilized exclusively for safety purposes—allowing the identification of complex, multi-request attacks and the refinement of false-positive triggers.
While Fable 5 is locked down for general safety, Claude Mythos 5 is being deployed through specialized channels. Initially available via Project Glasswing in collaboration with the US government, Mythos 5 strips away the cyber safeguards, granting elite cyberdefenders the ultimate tool for securing critical infrastructure. Soon, a similar trusted access program will open for biological researchers, allowing them to utilize Mythos 5 to accelerate biomedical discoveries safely.
Availability and Pricing Rollout
Both Fable 5 and Mythos 5 are aggressively priced at $10 per million input tokens and $50 per million output tokens, which is less than half the price of the previous Claude Mythos Preview.
With today’s launch, the era of Mythos-class AI has officially begun. By balancing unprecedented, industry-disrupting capabilities with cautious, intelligent safety systems, this release marks a mature step forward in the quest to bring advanced AI to the masses without compromising global security.
