Anthropic launched Claude Fable 5 on June 9, 2026, calling it the most capable model it has ever released to the general public. Alongside it, the company released Claude Mythos 5 for a small group of cybersecurity defenders and infrastructure providers — the same underlying model as Fable 5, but with some safeguards lifted. The shared root is deliberate: "Fable," from the Latin fabula, "that which is told," is akin to the Greek mythos. What separates the two is not the model but the guardrails around it.
For a reader trying to make sense of the launch, that is the whole story in miniature. Anthropic has shipped a system whose unrestricted twin it considers too dangerous to release openly — days after publicly warning that frontier AI is becoming dangerously capable — and the thing that made a broad release possible is not a weaker model but a stronger cage.
What Is a Mythos-Class Model?
Mythos-class models are a new tier Anthropic places above its Opus class in capability. The first, Claude Mythos Preview, launched in April through Project Glasswing to a limited group of cyber defenders and critical-infrastructure providers, held back because of its advanced cybersecurity abilities. Fable 5 is the version Anthropic judged safe enough for everyone; Mythos 5 is the upgraded restricted version, deployed initially through Glasswing in collaboration with the US government and described as having the strongest cybersecurity capabilities of any model in the world.
Pricing for both is $10 per million input tokens and $50 per million output tokens, less than half the cost of Mythos Preview. Fable 5 is available today through the Claude API and consumption-based Enterprise plans; for subscription tiers the rollout is staged, included free on Pro, Max, Team, and seat-based Enterprise plans through June 22, then moving to usage credits on June 23 until capacity expands, after which Anthropic intends to restore it as a standard plan feature.
How Capable Is Claude Fable 5?
By Anthropic's account, Fable 5 is state-of-the-art on nearly every tested benchmark, across software engineering, knowledge work, vision, and scientific research — and its lead grows the longer and more complex the task. In software engineering, Stripe reported during early testing that the model compressed months of work into days: in a 50-million-line Ruby codebase, it completed a codebase-wide migration in one day that would have taken a team more than two months by hand. On Cognition's FrontierCode evaluation, which checks whether a model can pass hard coding tasks while meeting production-quality standards, Fable 5 scored highest among frontier models even at medium effort, making it more token-efficient than past Claude models. Cursor's chief executive said it is the state of the art on the company's internal CursorBench and has opened up long-horizon problems that were previously out of reach.
In knowledge work, Fable 5 posted the top score on Hebbia's senior-level finance benchmark, with large gains in document reasoning and chart and table interpretation, while the trading firm IMC said it aced its trading-analysis evaluations nearly across the board. On vision — where Anthropic calls it the new state of the art — it can pull precise numbers from detailed scientific figures and rebuild a web app's source code from screenshots alone, with less scaffolding than before. The headline demonstration: earlier Claude models struggled to play Pokémon FireRed even with helper tools, but Fable 5 finished the game using only raw screenshots, with no maps or navigation aids. Given persistent file-based memory while playing the deck-builder Slay the Spire, it improved three times as much as Opus 4.8 and reached the final act three times as often.
What Did Mythos 5 Do in the Lab?
The most striking claims attach to Mythos 5, the less-restricted version used for internal science. Anthropic's protein-design experts said it sped up parts of the drug-design process roughly tenfold; in one test, equipped with protein-design and bioinformatics tools but no human help, it matched or beat skilled human operators, handling the full workflow a scientist normally owns — choosing binding sites, selecting and running design tools, and recovering from failures. Of 14 protein targets in that study, nine produced strong drug-design candidates now under investigation.
Anthropic also calls Mythos 5 its first model to consistently generate novel, compelling scientific hypotheses: in blinded comparisons against Opus-class models, its scientists preferred Mythos 5's molecular-biology hypotheses about 80% of the time, and one — a new mechanism for an E. coli protein — was corroborated by an independent lab. In genomics, it ran original research across more than a week of largely autonomous work, assembling single-cell data spanning 138 animal species and training a custom machine-learning model that outperformed a recently published Science model at one-hundredth the size. On alignment, Anthropic's automated assessment put Mythos 5's rate of misaligned behavior low and similar to Opus 4.8 — and since Fable 5 is the same underlying model, comparable.
How Do Fable 5's Safeguards Actually Work?
The mechanism that makes a broad release possible is a set of classifiers — separate AI systems that detect potential misuse — covering three areas. When a request trips one, the answer is handled by Anthropic's next-most-capable model, Claude Opus 4.8, rather than refused, and the user is told a handoff happened. Crucially, that design choice means a flagged request degrades to a still-capable model instead of a dead end, and Anthropic says more than 95% of Fable 5 sessions trigger no fallback at all — meaning for most users, performance is effectively identical to Mythos 5.
The first classifier covers cybersecurity, where Mythos-class models are unusually strong at finding software vulnerabilities and running multi-step attack tasks. Anthropic says it ran more than 1,000 hours of external bug-bounty testing without producing a universal jailbreak, that outside red teams found none on long-form agentic tasks so far — though the UK AI Safety Institute made some progress in a brief window — and that one external partner rated Fable 5's protection against harmful cyber queries the strongest of any model tested. The second covers biology and chemistry: Anthropic says it previously blocked only a narrow set of bioweapons-related queries but no longer considers that enough, citing both stronger real-world scientific capability and concern about well-resourced bad actors, so Fable 5 now routes most biology and chemistry requests to Opus 4.8. The third targets distillation — attempts to extract Claude's capabilities to train competing models — after the company said it identified large-scale extraction attempts tied to authoritarian countries.
What Changes for Data and Access?
Anthropic is also imposing 30-day retention on all traffic for Mythos-class models, across first- and third-party platforms. It says the data will not be used to train models or for any non-safety purpose, that it logs all human access, and that the data is deleted after 30 days in almost all cases — the aim being to defend against complex, multi-request attacks and to identify and reduce false positives. Access widens in stages: Glasswing partners can upgrade to Mythos 5 today with cyber safeguards lifted, and within weeks a trusted-access program will let vetted biomedical researchers use a version of the model with the biology and chemistry safeguards removed while cyber safeguards stay in place.
Why This Launch Matters
The most consequential part of Fable 5 may not be any single benchmark but the release model it establishes. Capability is no longer the thing that decides what ships; the safeguards are. Anthropic has packaged a frontier system for public use by gating it with deterministic classifiers and tiered trusted access rather than withholding it — fallback instead of refusal, mandatory retention, and restricted tiers for the most sensitive cyber and biology work. The practical catch for users is that conservatively tuned classifiers will sometimes reroute harmless requests to a less capable model, a friction Anthropic acknowledges and pledges to narrow. The wider significance is that this approach now reads as a template other frontier labs facing the same dual-use dilemma may be pushed to follow.
Frequently Asked Questions
What is the difference between Claude Fable 5 and Mythos 5?
They are the same underlying model. Fable 5 is the version released to the general public with safety classifiers active; Mythos 5 has some of those safeguards lifted and is restricted to vetted cybersecurity partners through Project Glasswing, with a biology trusted-access program planned. The names reflect the safeguards, not a difference in capability.
How much does Claude Fable 5 cost?
Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens — less than half the cost of Claude Mythos Preview. On subscription plans, Fable 5 is included free through June 22, 2026, then requires usage credits until Anthropic expands capacity.
What do Claude Fable 5's safeguards do?
When Fable 5's classifiers detect a request related to cybersecurity, biology and chemistry, or model distillation, the response is handled by Claude Opus 4.8 instead of Fable 5, and the user is notified. Anthropic says this fallback occurs in fewer than 5% of sessions, so most users get Fable 5's full capability.
Is Claude Fable 5 better than Claude Opus 4.8?
Anthropic reports Fable 5 is state-of-the-art on nearly all tested benchmarks and beats Opus 4.8 on long, complex tasks, with early customers reporting double-digit gains on some evaluations. The biggest improvements show up on long-horizon agentic work in coding, analysis, and scientific research.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
