By Vijeth Shivappa
Mythos-class models represent a new tier of Claude models that sit above the Opus class in capability. The first in this lineage, Claude Mythos Preview, was released in April through Project Glasswing. Claude Fable 5 and Mythos 5 now follow as the next step.
The naming is deliberate and layered. Fable derives from the Latin fabula, meaning “that which is told,” and is closely related to the Greek mythos. The safeguards are what distinguish the two models and explain why they carry different names. This is not just branding; the naming encodes the philosophy of the release: the same underlying intelligence, with different access gates.
This new tier reflects a maturing AI landscape in which models are no longer judged only by how much smarter they are, but by how well they can handle long-horizon cognitive labor.
The Two-Model Strategy: Public vs. Restricted
The launch is structured around a deliberate bifurcation.
Claude Fable 5 is the first publicly available Mythos-class model. Claude Mythos 5 is the same underlying model with safeguards removed, available only in a limited program, Project Glasswing, for infrastructure providers and vetted cybersecurity researchers. In practice, when people refer to the new Mythos model, the version developers can actually access is Fable 5.
Claude Fable 5 is generally available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry. Claude Mythos 5 is offered in limited availability only to approved customers in Project Glasswing. Customers without Mythos 5 access can use Fable 5, which offers the same capabilities.
This is a philosophically interesting position: Anthropic is essentially saying that the ceiling of human-accessible AI capability is now gated not by the model’s intelligence, but by the safety layer around it.
Capabilities: What Makes This a Generational Leap
Software Engineering
Stripe reported that Fable 5 compressed months of engineering work into days. In a 50-million-line Ruby codebase, the model performed a codebase-wide migration in a single day, a task that would otherwise have taken a whole team more than two months to complete manually.
On Cognition’s FrontierCode evaluation, which tests whether models can solve difficult coding tasks while meeting the standards of high-quality production codebases, Fable 5 scored highest among frontier models, even at medium effort.
As tasks grow in duration and complexity, the performance gap between Fable 5 and its predecessors widens significantly, suggesting that Mythos-tier models may be becoming the new baseline for high-level autonomous work.
Knowledge Work and Finance
On Hebbia’s Finance Benchmark for senior-level reasoning, Fable 5 achieved the highest score of any model, with substantial gains in document-based reasoning, chart and table interpretation, and problem solving. IMC noted that Fable 5 performed exceptionally well on trading-analysis evaluations across factual lookup, conceptual reasoning, root-cause analysis, and expected-value analysis.
Vision
Fable 5 is now the new state-of-the-art model for vision tasks. It can extract precise numbers from detailed scientific figures and rebuild a web app’s source code from screenshots alone. Previous Claude models struggled to play Pokémon FireRed even with harnesses that gave them additional tools; Fable 5 beat FireRed with a minimal, vision-only harness.
Memory and Long-Context Performance
Fable 5 stays focused across millions of tokens in long-running tasks and improves its outputs using its own notes. In tests on the deck-building game Slay the Spire, giving the model access to persistent file-based memory improved its performance three times more than it did for Opus 4.8, and Fable also reached the game’s final act three times more often.
Life Sciences and Drug Design
This is arguably the most consequential domain.
Using Mythos 5, internal protein-design experts accelerated parts of the drug-design process by around ten times. Mythos 5, equipped with protein-design and bioinformatics tools but no human assistance, matched or beat skilled human operators by completing the tasks normally handled by a scientist: choosing binding sites, selecting and running protein-design tools, and recovering from failures along the way. Nine of 14 protein targets yielded strong drug-design candidates that are now under investigation.
Mythos 5 is also Anthropic’s first model to consistently produce novel, compelling scientific hypotheses. In blinded head-to-head comparisons, scientists preferred Mythos’s molecular biology hypotheses roughly 80% of the time, and several have advanced to experimental evaluation. One Mythos hypothesis — a novel mechanism for an E. coli protein — was later corroborated in an independent study from a lab working on the same problem.
The Safety Architecture: “Graceful Degradation”
The most structurally novel aspect of this release is what Anthropic calls its fallback-to-Opus approach. Rather than issuing flat refusals, when Fable 5’s classifiers detect a sensitive query, the response is automatically handled by Claude Opus 4.8 instead. Users are informed whenever this occurs. Opus 4.8 is highly capable in its own right, so a fallback response is far better than an outright refusal. Early data shows that more than 95% of Fable sessions involve no fallback at all.
Three domains are covered by these classifiers.
1. Cybersecurity
Mythos-class models excel at discovering and exploiting software vulnerabilities, making cyberattacks substantially easier and cheaper to commit. The classifiers cover both exploitation and offensive cyber tasks more broadly, preventing Fable from making any progress on agentic hacking tasks.
2. Biology and Chemistry
Mythos 5 was tested on its ability to complete a challenging step in designing adeno-associated viruses (AAVs), which are used to deliver gene therapies but could also, in the wrong hands, enable dangerous virus design. Mythos-class models outperformed sophisticated protein language models using biological reasoning alone, highlighting the dual-use risk.
3. Distillation
Anthropic has previously identified large-scale attempts to extract Claude’s capabilities and train competing models in authoritarian countries. Distilling Fable 5’s abilities could help spread near-frontier AI capabilities without adequate safeguards. Requests flagged as distillation attempts fall back to Opus 4.8.
Jailbreak Resistance
Internally, Anthropic ran an external bug bounty that produced no universal jailbreaks over more than 1,000 hours of testing. External red-teaming organizations also failed to find universal jailbreaks on long-form agentic tasks, although the UK AISI made progress toward one within a brief initial testing window.
One external partner found Fable 5’s safeguards against harmful cyber queries to be the most robust of any model tested, including Opus 4.8 and 4.7. Fable 5 complied with zero harmful single-turn requests related to planning a cyberattack, exploit development, or defense evasion, whether or not any of 30 public jailbreak techniques were used.
Alignment Assessment
In Anthropic’s automated alignment assessment, Mythos 5’s level of misaligned behavior — including deception and cooperation with misuse by a user — was low and similar to that of Opus 4.8. Because Fable 5 and Mythos 5 are the same underlying model, Fable 5’s alignment profile is comparable.
Pricing and Availability
Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, exactly double Opus 4.8. Paid Claude subscribers get free access during the June 9–22, 2026 introductory window.
Both models carry a 30-day data retention policy and are not available under zero data retention. Anthropic says the data will not be used for model training, but will help defend against complex and novel attacks, including new jailbreaks and coordinated attacks across many requests.
Strategic Context: What This Launch Signals
Several macro signals stand out for enterprise and policy observers.
Capability Ceiling Is Shifting
The fact that a model Anthropic deemed too dangerous for public release in April is now available to all developers by June — with the same underlying weights — suggests that safety tooling is now moving almost as quickly as capability growth.
The Trusted-Access Tier Is Expanding
Anthropic plans to open a trusted access program for biology, providing access to Fable 5 with biology and chemistry safeguards removed but cyber safeguards retained, and enrolling life science organizations across fundamental and translational research.
Recursive Self-Improvement Proximity
The launch follows Anthropic’s public warning that frontier AI is advancing so rapidly that systems may soon achieve recursive self-improvement, meaning they could improve themselves without human intervention. Fable 5 is being released against a backdrop of Anthropic’s own concern about the pace of development.
IPO Timing
Fable’s launch comes as Anthropic prepares to enter the public markets, alongside OpenAI and Elon Musk’s SpaceX, giving the release both technical and commercial significance.
Summary Assessment
Claude Fable 5 and Mythos 5 represent Anthropic’s most significant capability release to date, not just in raw performance but in the philosophical and architectural approach to deploying frontier AI. The graceful-degradation safety model, the two-tier public/restricted access structure, and the life sciences breakthroughs together mark a new chapter.
For enterprise practitioners in sectors like BFSI, healthcare, and critical infrastructure, this is the model tier that begins to make agentic AI genuinely production-grade, with compliance and safety scaffolding starting to catch up to the capability curve.
