Anthropic Releases Claude Fable 5 and Claude Mythos 5: Same Underlying Model, Different Safeguards, New Mythos-Class Tier

Anthropic has released two new AI models, Claude Fable 5 and Claude Mythos 5, both falling under the new "Mythos-class" tier. This tier surpasses the previous Opus class in capability. Fable 5 is engineered for general use with comprehensive safety measures, while Mythos 5, though based on the same core model, has certain safeguards relaxed, making it available for limited, specific applications. The names "Fable" (from Latin "fabula," that which is told) and "Mythos" (Greek for the same) reflect their distinct applications and safety profiles. Both models support a 1M token context window and allow up to 128k output tokens per request, with competitive pricing significantly lower than prior versions.

Fable 5 is positioned as Anthropic's most capable widely released model, excelling in demanding reasoning and long-horizon agentic tasks. It has demonstrated state-of-the-art performance across software engineering, knowledge work, vision, and scientific research. For instance, Stripe utilized Fable 5 for a codebase-wide migration of a 50-million-line Ruby codebase, completing in one day what would typically take a team over two months. Its efficiency is also notably improved, scoring highest in frontier model evaluations for difficult coding tasks.

In knowledge work, Fable 5 achieved the highest score on Hebbia's Finance Benchmark for senior-level reasoning, showing significant gains in document-based reasoning, chart and table interpretation, and problem-solving. For vision tasks, it can extract precise numbers from complex scientific figures and rebuild web application source code from screenshots. Furthermore, Fable 5 demonstrates superior memory and long-context capabilities, staying focused across millions of tokens and improving outputs using self-generated notes.

Mythos 5, the less safeguarded version, is geared towards scientific advancements. Internal testing showed it accelerated parts of drug design tenfold and consistently generated novel scientific hypotheses, preferred by scientists in blinded comparisons. It also conducted novel genomics research autonomously, training a custom model that outperformed a published Science model despite being 100 times smaller.

The deployment of such powerful models carries inherent risks. To mitigate potential misuse, especially in cybersecurity, Fable 5 incorporates a new set of classifiers—separate AI systems designed to detect and prevent malicious requests, including jailbreak attempts. When a request is flagged, the response is handled by Claude Opus 4.8, and users are informed of this fallback. Anthropic has conservatively tuned these safeguards, resulting in fallbacks in less than 5% of sessions, and extensively red-teamed the classifiers to ensure their robustness.

Mythos 5, with its lifted cyber safeguards, boasts the strongest cybersecurity capabilities among current models. It is deployed through Project Glasswing in collaboration with the U.S. government, providing technical teams with advanced tools for various concrete workflows. Additionally, Anthropic plans a trusted access program for Mythos 5 in biology, offering approved researchers access to its full capabilities without specific biological safeguards.