Anthropic Claude Fable 5 & Mythos 5: Features, Release Updates 2026

Why This Launch Is Different From Previous Releases

Anthropic has made capable models before. What changed with Fable 5 and Mythos 5 is the scale of what's being released and the deliberate structure around it.

Claude Mythos Preview, the first Mythos-class model, launched in April 2026 through Project Glasswing, a restricted program for cybersecurity organizations and critical infrastructure providers. Access was intentionally narrow because the model's capabilities particularly around finding software vulnerabilities were considered too risky for a public launch.

Two months later, Anthropic found a way to release Mythos-class performance more broadly. The solution was to build a separate model with safeguards built in: Claude Fable 5. It runs the same underlying architecture as Mythos 5, but with classifiers that intercept sensitive queries and route them to Claude Opus 4.8 instead. The result is a generally available model that, in over 95% of sessions, performs identically to Mythos 5.

The name "Fable" comes from the Latin fabula, meaning "that which is told" — a deliberate nod to its Greek counterpart, mythos. The names signal that the two models are related but distinct, and the safeguards are what separates them.

Claude Fable 5: What It Can Do

Fable 5 is Anthropic's most capable model to date for general use. Its advantage grows with task complexity and length, and early partner results put it ahead of other frontier models across several categories.

In software engineering, Stripe found that Fable 5 compressed months of work into days during early testing. In a 50-million-line Ruby codebase, the model completed a full migration in a single day work that would have taken an entire team more than two months by hand. On Cognition's FrontierCode benchmark, which tests whether models can pass difficult coding tasks at production-quality standards, Fable 5 ranked first among frontier models.

For knowledge work, results from Hebbia's Finance Benchmark placed Fable 5 above all other models on senior-level reasoning tasks, with strong improvements in document analysis, chart interpretation, and structured problem solving. IMC reported that the model handled their trading analysis evaluations across the board, including factual lookup, root-cause analysis, and expected-value calculations.

On vision tasks, Fable 5 sets a new bar. It extracts precise values from detailed scientific charts, can reconstruct a web application's source code from screenshots alone, and needs less scaffolding than previous models. Earlier Claude models required additional tools to navigate games like Pokémon FireRed; Fable 5 completed it using only raw game screenshots.

For long-context work, the model maintains coherence across millions of tokens and can use persistent notes to improve its own outputs mid-task. In tests using the deck-building game Slay the Spire, Fable 5 benefited from memory access three times more than Opus 4.8, and reached the game's final act three times more often.

Claude Mythos 5: The Restricted Version

Claude Mythos 5 is the same underlying model as Fable 5, with safety classifiers removed in specific areas. It currently serves cybersecurity organizations through Project Glasswing and, in the coming weeks, a small group of biomedical researchers through a new trusted access program.

In cybersecurity, Anthropic describes Mythos 5 as holding the strongest vulnerability-finding capabilities of any model currently available. Within Project Glasswing, it has already been used to help defenders secure critical software infrastructure.

In life sciences, the results from internal testing are notable. Protein design experts used Mythos 5 to accelerate drug design workflows by around ten times. The model handled a complete scientific workflow without human assistance choosing binding sites, selecting tools, and recovering from failures and matched or outperformed skilled human operators. Nine of 14 protein targets from that work yielded strong drug design candidates currently under further investigation.

In molecular biology, Anthropic's scientists preferred Mythos 5's hypotheses over Opus-class outputs roughly 80% of the time in blinded comparisons. One hypothesis about an E. coli protein mechanism was independently corroborated by a separate research lab working on the same problem.

Mythos 5 also conducted over a week of largely autonomous genomics research assembling single-cell data across 138 animal species and training a custom machine learning model that outperformed a recently published model in the journal Science, despite being 100 times smaller.

How the Safeguards Work

Fable 5's classifiers are separate AI systems running alongside the model. When a query is flagged, the response is handled by Claude Opus 4.8 rather than Fable 5, and users are informed when this happens. Anthropic notes that Opus 4.8 is a strong model in its own right, so the fallback is meaningfully better than a flat refusal.

Three areas trigger the classifiers. The first is cybersecurity: Mythos-class models can discover and exploit software vulnerabilities, execute reconnaissance, and perform multi-stage offensive operations. Because the same capability is valuable to defenders and dangerous in other hands, the classifiers cover both direct exploitation and offensive tasks broadly. External testing, including a bug bounty program with over 1,000 hours of work, found no universal jailbreaks. An external partner found Fable 5's cyber safeguards the most robust of any model tested with zero compliance on harmful requests across 30 different public jailbreak techniques.

The second area is biology and chemistry. Mythos-class models can now complete real scientific tasks that previously required specialized tools. In one evaluation, Mythos 5 outperformed dedicated protein language models at predicting properties of adeno-associated virus shells relevant to gene therapy but also to potential misuse. Fable 5 currently falls back to Opus 4.8 on most biology and chemistry queries while Anthropic works to narrow those safeguards.

The third is distillation. Anthropic has identified large-scale attempts to extract Claude's capabilities to train competing models without appropriate safeguards. Queries flagged as distillation attempts are also routed to Opus 4.8. A new data retention policy accompanies the launch: 30-day retention for all Mythos-class traffic on first- and third-party surfaces. The data is not used for training, access is logged, and deletion happens after 30 days in almost all cases.

Practical Use Cases for Teams and Developers

For software teams, Fable 5 is available now through the Claude API using the model string claude-fable-5. It handles large-scale code migrations, production-level code reviews, pull request summaries, debugging, and documentation generation across complex codebases.

For knowledge workers and analysts, the model's performance on document reasoning and structured analysis makes it applicable to financial research, contract review, and report generation at scale.

For developers building AI-powered products, Fable 5's vision capabilities and long-context memory open up use cases in interface reconstruction, scientific data extraction, and multi-step autonomous workflows.

For cybersecurity and research organizations, Mythos 5 access currently runs through Project Glasswing. A broader trusted access program is planned, and biomedical researchers will be able to apply for biology-focused access in the coming weeks.

What This Signals About Where AI Is Heading

The Fable/Mythos structure reflects a question the industry hasn't fully resolved: how do you release genuinely powerful AI without either locking it away entirely or releasing it without control? Anthropic's answer here is tiered access with transparent safeguards, a capable fallback model rather than outright refusals, and a clear expansion path as the safety work matures.

For most users and organizations, Fable 5 is the relevant development — and it's available today. For those in cybersecurity or life sciences research, Mythos 5 is the more significant model, and access to it will expand over the coming months.

Chat & Ask AI brings together multiple advanced models for writing, research, coding, document analysis, and everyday tasks. For users who want capable AI assistance without switching between platforms, it offers a single environment to explore what modern AI can handle.