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

Anthropic launched two fashions on June 9, 2026: Claude Fable 5 and Claude Mythos 5. Both belong to a tier referred to as “Mythos-class.” This tier sits above the Opus class in functionality. Fable 5 is the model claimed to be made secure for basic use. Mythos 5 is similar mannequin with some safeguards lifted, stored in restricted launch.

Claude Fable 5 and Mythos 5

Mythos-class fashions are a tier of Claude fashions. They sit above the Opus class in functionality. The first was Claude Mythos Preview, launched in April via Project Glasswing.

Fable 5 and Mythos 5 share the identical underlying mannequin. The distinction is the safeguards. Fable 5 ships with security classifiers for basic use. Mythos 5 has some classifiers eliminated and stays in restricted launch.

The names mirror this break up. “Fable” comes from the Latin fabula, “that which is instructed.” This is akin to the Greek mythos. The safeguards distinguish the 2 fashions, in order that they carry totally different names.

Anthropic crew calls Fable 5 its most succesful broadly launched mannequin. It targets demanding reasoning and long-horizon agentic work. Anthropic states Fable 5’s capabilities exceed any mannequin it has made typically out there.

Both fashions help a 1M token context window by default. They permit as much as 128k output tokens per request. Pricing is $10 per million enter tokens and $50 per million output tokens. That is lower than half the worth of Claude Mythos Preview.

The Capability Case

Anthropic reviews Fable 5 is state-of-the-art on practically all examined functionality benchmarks. It reveals robust outcomes throughout software program engineering, data work, imaginative and prescient, and scientific analysis. The longer and extra advanced the duty, the bigger its lead over Anthropic’s different fashions.

On software program engineering, Stripe examined Fable 5 throughout early entry. The mannequin carried out a codebase-wide migration in a 50-million-line Ruby codebase. According to Stripe: this took sooner or later. By hand, a crew would have wanted over two months.

Fable 5 can be extra token-efficient than previous Claude fashions. On Cognition’s FrontierCode analysis, Fable 5 scores highest amongst frontier fashions. This holds even at medium effort. The eval assessments tough coding duties beneath production-codebase requirements.

On data work, Anthropic cites Hebbia’s Finance Benchmark for senior-level reasoning. Fable 5 posts the very best rating of any mannequin there. Gains are available document-based reasoning, chart and desk interpretation, and downside fixing.

On imaginative and prescient, Anthropic calls Fable 5 the brand new state-of-the-art. It can extract exact numbers from detailed scientific figures. It can rebuild an internet app’s supply code from screenshots alone. It additionally wants much less scaffolding than prior fashions. Fable 5 beat Pokémon FireRed with a minimal, vision-only harness.

On reminiscence and long-context, Fable 5 stays targeted throughout hundreds of thousands of tokens. It improves its outputs utilizing its personal notes. In the sport Slay the Spire, persistent file-based reminiscence helped it 3 times greater than Opus 4.8.

Mythos 5 carries the science claims. Internal protein design specialists accelerated elements of drug design by round ten occasions. Anthropic additionally says Mythos 5 is its first mannequin to constantly produce novel scientific hypotheses. Scientists most popular its molecular biology hypotheses round 80% of the time in blinded comparisons.

Mythos 5 additionally ran novel genomics analysis over per week of largely autonomous work. It educated a customized mannequin on single-cell knowledge spanning 138 animal species. Anthropic says that mannequin outperformed a latest mannequin revealed in Science, regardless of being 100 occasions smaller.

How the Safeguards Work

Releasing a mannequin this succesful carries threat. Without safeguards, Fable 5’s cybersecurity capabilities may very well be misused to trigger critical harm. Anthropic due to this fact launched Fable 5 with a brand new set of classifiers.

Classifiers are separate AI techniques. They detect potential misuse, together with jailbreak makes an attempt. They stop the primary mannequin from responding to flagged requests.

When Fable 5’s classifiers flag a request, the response is dealt with by Claude Opus 4.8 as a substitute. The coated areas are cybersecurity, biology and chemistry, and distillation. Users are knowledgeable every time a fallback happens.

For biology and chemistry, Fable 5 falls again to Opus 4.8 on most requests for now. Anthropic cites concern that the identical dual-use queries might give uplift to malicious actors. It plans a trusted entry program for biology, giving authorized researchers Fable 5 with out these safeguards.

Anthropic tuned these safeguards conservatively. They will typically catch innocent requests. On common, they set off in lower than 5% of periods. Anthropic says greater than 95% of Fable periods contain no fallback in any respect. For these periods, Fable 5’s efficiency successfully matches Mythos 5.

Anthropic red-teamed the classifiers extensively. An exterior bug bounty produced no common jailbreaks in over 1,000 hours. A common jailbreak lets a person work together with the mannequin as if its safeguards had been absent. Anthropic notes the UK AISI made progress towards one in a short testing window.

Mythos 5 is similar mannequin with cyber safeguards lifted. Anthropic describes it as having the strongest cybersecurity capabilities of any present mannequin. It is deployed via Project Glasswing in collaboration with the US authorities.

Use Cases

These capabilities map to a number of concrete workflows for technical groups:

• Large-scale code migration: Long-horizon coding fits large refactors and cross-repo migrations. The Stripe instance reveals this at a 50-million-line scale.
• Agentic coding pipelines: Fewer turns and token effectivity assist multi-step agent runs. GitHub reported autonomy and reliability on advanced, long-horizon coding duties.
• Finance and analytics work: Strong doc and chart reasoning fits senior-level monetary evaluation. Hebbia and IMC cited features on reasoning and trading-analysis duties.
• Vision-to-code duties: Rebuilding supply from screenshots fits front-end reconstruction and determine extraction. The vision-only harness reduces tooling overhead.
• Long-running analysis brokers: Persistent reminiscence throughout hundreds of thousands of tokens fits multi-day analysis loops. Mythos 5 ran novel genomics work over per week of largely autonomous work.

(*5*)Comparison Table: Fable 5 vs. Mythos 5 vs. Opus 4.8

Attribute	Claude Fable 5	Claude Mythos 5	Claude Opus 4.8
Model tier	Mythos-class	Mythos-class	Opus class
Underlying mannequin	Same as Mythos 5	Same as Fable 5	Opus 4.8
Availability	Generally out there	Limited (Project Glasswing)	Generally out there
Safety classifiers	Active (cyber, bio/chem, distillation)	Cyber safeguards lifted	Opus-level safeguards
Fallback goal	Falls again to Opus 4.8	Not relevant	Not relevant
API mannequin ID	claude-fable-5	claude-mythos-5	(present Opus ID)
Context window	1M tokens default	1M tokens default	Per Opus specs
Max output	128k tokens/request	128k tokens/request	Per Opus specs
Input worth (per 1M)	$10	$10	(per Opus pricing)
Output worth (per 1M)	$50	$50	(per Opus pricing)
Thinking mode	Adaptive solely, all the time on	Adaptive solely, all the time on	Configurable
Data retention	30-day (Covered Model)	30-day (Covered Model)	Standard choices

Note: Specific Opus 4.8 specs and pricing will not be detailed within the Fable 5 launch sources. The desk marks these cells accordingly.

---

Key Takeaways

• Fable 5 and Mythos 5 share one underlying mannequin; safeguards are the one distinction.
• Anthropic reviews Fable 5 is state-of-the-art on practically all examined functionality benchmarks.
• Fable 5 classifiers fall again to Opus 4.8 and set off in beneath 5% of periods.
• Both fashions provide a 1M token context window at $10 enter and $50 output per million tokens.
• Mythos 5 stays restricted to Project Glasswing; Fable 5 is mostly out there throughout main platforms.

Sentiments of the Community