Anthropic just shook the AI world again. Meet Claude Fable 5—its first Mythos-class model released to the public. This isn’t a minor upgrade. It’s a leap above their previous top-tier, Opus 4.8.

Fable 5 and its sibling, Mythos 5, share the exact same core engine. The only difference: safety guards. Fable 5 wears a strict safety layer that blocks risky queries. Mythos 5 drops some of those limits but stays locked behind a vetted-access wall.

The Mythos class sits above Anthropic’s earlier tiers—Haiku, Sonnet, and Opus. Mythos models boast a massive 1 million token context window and can output up to 128,000 tokens in one go. That’s a playground for long, complex tasks.

Benchmarks don’t lie. Fable 5 dominates software engineering tests, knowledge work, vision tasks, and scientific research. It scored 80.3% on the demanding SWE-Bench Pro, outpacing Opus 4.8 by over 11 points. It even crushed GPT-5.5 and Gemini 3.1 Pro, models considered cutting-edge themselves.

Real-world feats back the hype. Stripe used Fable 5 to migrate a 50-million-line Ruby codebase in a single day. Humans would need two months. Enterprise feedback highlights Fable’s ability to autonomously plan, verify, and fix multi-step jobs—making it a reliable agent not just a fancy autocomplete.

Vision-wise, Fable 5 can extract precise data from scientific charts and rebuild web app code from screenshots alone. In gaming, it beat Pokémon FireRed using only vision cues. It even demonstrated persistent memory, tripling Opus 4.8’s performance in long-context tasks like the game Slay the Spire.

Mythos 5 takes the science claims further. It helped internal protein designers speed up drug development by ten times. It autonomously generated novel molecular biology hypotheses, preferred by scientists 80% of the time. Over a week, Mythos 5 ran near-autonomous genomics research, training a model on single-cell data from 138 animal species that outperformed a much larger recent Science-published model.

Safety is the headline here. Anthropic knows Mythos power can get dangerous. Without safeguards, Fable 5’s cybersecurity skills could be exploited for harm. So it wraps Fable 5 in AI classifiers that detect misuse and jailbreak attempts. When flagged, queries fall back to Opus 4.8, a less capable but safer model.

These safeguards cover cybersecurity, biology, chemistry, and distillation domains. They trigger in less than 5% of sessions, meaning 95% of the time users get full Mythos power. Anthropic even ran 1,000+ hours of external bug bounties—no universal jailbreak found. They did note a brief breakthrough by UK’s AISI, but no lasting exploit.

Mythos 5, with lifted cyber safeguards, runs only in specialized partnerships, including US government programs and select biology researchers. Anthropic plans to expand trusted access for these fields but keeps Mythos 5 tightly controlled due to risk.

Price tags reflect power and safety. Fable 5 and Mythos 5 cost $10 per million input tokens and $50 per million output tokens—double Opus 4.8’s rates but less than half of the Mythos Preview’s initial cost. Anthropic offers batch pricing to soften costs for high-volume users.

Fable 5 is now available through Claude’s API and enterprise plans. Subscription access rolled out gradually, free within certain plans until June 22, then shifting to usage credits. Enterprises wary of AI expenses will have to weigh Fable’s autonomy gains against steeper bills.

One new policy may raise eyebrows: Anthropic enforces a 30-day data retention on all traffic for Fable and Mythos models, even for enterprises with previous zero-retention agreements. They insist this data won’t train models but will help spot new attacks and reduce false positives—a tradeoff between power and privacy.

Anthropic’s Mythos-class launch signals a new frontier for AI. It’s not just raw capability anymore. It’s about delivering that power with guardrails strong enough to prevent misuse. Fable 5 embodies that balance—pushing boundaries while keeping them in check. Whether the industry follows suit remains to be seen.