On June 9, 2026, Anthropic launched Claude Fable 5 — a Mythos-class model made safe for general use — alongside Claude Mythos 5 for a limited group of cyber defenders and infrastructure providers. Fable 5 exceeds every model Anthropic has previously made generally available and leads on nearly all tested benchmarks, with its advantage growing as tasks become longer and more complex.
Releasing a model this capable carries real risks. Without safeguards, Fable 5's cybersecurity abilities could be misused to cause serious harm. Anthropic therefore ships the public model with safety classifiers that route some queries to Claude Opus 4.8 instead. These safeguards are tuned conservatively — they sometimes catch harmless requests, but trigger in fewer than 5% of sessions on average.
Claude Mythos 5 uses the same underlying weights as Fable 5 but with safeguards lifted in selected areas. It is initially deployed through Project Glasswing in collaboration with the U.S. government, upgrading Claude Mythos Preview. Anthropic calls it the strongest cybersecurity model in the world and plans to expand access through a broader trusted-access program.
Both models are priced at $10 per million input tokens and $50 per million output tokens — less than half the price of Claude Mythos Preview.
Evaluating Claude Fable 5 and Claude Mythos 5
Fable 5 and Mythos 5 can work autonomously longer than any previous Claude model. Anthropic highlights advances in software engineering, knowledge work, vision, memory, and life-sciences research.
Software engineering
Stripe reported that Fable 5 compressed months of engineering into days. In a 50-million-line Ruby codebase, the model completed a codebase-wide migration in one day — work estimated at over two months for a full team. On Cognition's FrontierCode evaluation, Fable 5 scores highest among frontier models even at medium effort, while using fewer tokens than prior Claude models.
Knowledge work
On Hebbia's Finance Benchmark for senior-level reasoning, Fable 5 leads all tested models with substantial gains in document reasoning, chart and table interpretation, and problem solving. IMC noted that Fable 5 aced trading-analysis evaluations across factual lookup, root-cause analysis, and expected-value analysis.
Vision
Fable 5 is state-of-the-art on vision tasks: extracting precise numbers from scientific figures and rebuilding a web app's source code from screenshots alone. Earlier Claude models needed complex helper harnesses to play Pokémon FireRed; Fable 5 completed the game with a minimal vision-only harness using raw screenshots.
Memory and long context
Fable 5 stays focused across millions of tokens in long-running tasks and improves outputs using its own notes. In deck-building game Slay the Spire, persistent file-based memory improved Fable 5's performance three times more than Opus 4.8; Fable also reached the final act three times more often.
Demonstrations shared by Anthropic include: a solar-system simulation derived from physics first principles to predict eclipses; autonomous Factorio gameplay; a browser-based CAD editor (built by Fable 5) designing 3D-printable models; and a fluid simulation synchronized to music Fable 5 generated in code.
Drug design and life sciences
Using Mythos 5, Anthropic's protein-design experts accelerated parts of the drug-design process by roughly ten times. In one study, Mythos 5 — with protein-design and bioinformatics tools but no human assistance — matched or beat skilled human operators across tasks normally done by scientists: choosing binding sites, running design tools, and recovering from failures. Nine of 14 protein targets yielded strong drug-design candidates under investigation.
Mythos 5 is Anthropic's first model to consistently produce novel, compelling scientific hypotheses. In blinded comparisons, scientists preferred Mythos molecular-biology hypotheses ~80% of the time over Opus-class outputs; several advanced to experimental evaluation. One hypothesis about an E. coli protein was independently corroborated by an external lab.
Mythos 5 also conducted novel genomics research over more than a week of largely autonomous work: assembling single-cell data for millions of cells across 138 animal species and training a custom model to identify analogous cell types across distant species. With only high-level human input, the model outperformed a recent Science journal publication despite being 100 times smaller.
Alignment
Anthropic's automated alignment assessment found Mythos 5's level of misaligned behavior — including deception and cooperation with misuse — low and similar to Opus 4.8. Because Fable 5 shares the same underlying model, its alignment profile is expected to be similar. Full details appear in the model system card.
Early feedback for Claude Fable 5
“Claude Fable 5 is the state of the art model on CursorBench. It's opened up a class of long-horizon problems that were out of reach for earlier models.”
“Claude Fable 5 is a real step forward for the developers GitHub serves. In our early testing, it took on complex, long-horizon coding tasks with a level of autonomy and reliability that exceeded previous benchmarks.”
“Claude Fable 5's reasoning is a clear step beyond Opus 4.8. It works at senior research scientist grade — picking directions, allocating resources, killing its incorrect beliefs, and producing novel first-principles outputs.”
“Claude Fable 5 is the first to break 90% on our core analytics benchmark of complex, long-running analytical tasks — a 10-point jump over Opus.”
“Claude Fable 5 is the strongest model we've tested on frontier physics research while using a third of the reasoning tokens.”
“Claude Fable 5 beats Opus 4.8 on our everyday spreadsheet suite at every effort level — and it does it with fewer turns, finishing runs 25–30% faster.”
Claude Fable 5's new safeguards
Mythos-class models present significant misuse risks. Since April, Anthropic has run Project Glasswing with Claude Mythos Preview for a limited group of cyber defenders. After months of improvement, safeguards are now robust enough for general release — though still deliberately cautious, with known false positives Anthropic aims to reduce post-launch.
Safety classifiers
Fable 5 ships with new classifiers: separate AI systems that detect potential misuse and jailbreak attempts. When classifiers flag requests related to cybersecurity, biology and chemistry, or model distillation, responses are handled by Claude Opus 4.8 instead. Users are informed when fallback occurs. More than 95% of Fable sessions involve no fallback at all.
Cybersecurity: Mythos-class models excel at finding and exploiting vulnerabilities and performing agentic hacking (reconnaissance, lateral movement, and more). Anthropic red-teamed classifiers extensively; an external bug bounty produced no universal jailbreaks in 1,000+ hours. One partner found Fable 5 complied with zero harmful single-turn cyberattack requests, including under 30 public jailbreak techniques.
Biology and chemistry: Classifiers currently fall back to Opus 4.8 on most biology and chemistry requests — broader than ideal, but chosen to enable a safe launch. Mythos 5 outperformed dedicated protein language models on an unpublished AAV shell-assembly prediction task, illustrating dual-use scientific capability. A trusted biology access program for Mythos 5 is planned.
Distillation: Requests flagged as large-scale capability extraction attempts — such as those Anthropic has observed from authoritarian countries — also fall back to Opus 4.8.
New data retention policy
For Fable 5, Mythos 5, and future models of similar capability, Anthropic requires 30-day retention for all traffic on Mythos-class models across first- and third-party surfaces. Data will not train new Claude models or be used for non-safety purposes; human access is logged and data is deleted after 30 days in almost all cases. Retention supports defense against novel jailbreaks and reducing false positives.
Claude Mythos 5 and the trusted access program
Glasswing partners with Mythos Preview access can upgrade to Mythos 5 today — same model as Fable 5 with cyber safeguards lifted, comparable or stronger than Mythos Preview at substantially lower cost. Anthropic plans to expand access in consultation with the U.S. government, including a systematic application process for cybersecurity organizations.
A biology trusted-access program will provide Fable 5 with biology and chemistry safeguards removed (cyber safeguards still active) to a small cohort of life-science researchers, with expansion planned alongside improved classifiers.
Availability and pricing
Claude Fable 5 is available everywhere today via the Claude API (model string: claude-fable-5), Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Claude Mythos 5 remains restricted to Glasswing partners and soon select biology researchers until broader trusted access opens.
- API and consumption-based Enterprise: Fable 5 fully available from launch day.
- June 9–22, 2026: Fable 5 included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost.
- From June 23: Fable 5 removed from those subscription tiers; continued use requires usage credits (Anthropic may extend the free window if capacity allows).
- Anthropic aims to restore Fable 5 as a standard subscription inclusion when capacity permits.
Fable 5 counts as 2× usage on subscription plans due to higher compute requirements. Anthropic will communicate rollout changes ahead of time.
Naming note
Mythos-class models sit above Opus in capability. Fable comes from the Latin fabula ("that which is told"), akin to the Greek mythos. The safeguards — not the weights — are what distinguish Fable 5 from Mythos 5.