Claude Fable 5 and Claude Mythos 5
Anthropic released two models on June 9 2026: Claude Fable 5 and Claude Mythos 5. Both belong to the “Mythos‑class” tier, which sits above the Opus class in capability. Fable 5 is marketed as the safe‑for‑general‑use version, while Mythos 5 is the same model with some safeguards lifted and is kept in a limited release.
The capability case
Anthropic reports that Fable 5 is state‑of‑the‑art on almost all tested benchmarks. Highlights include:
- Software engineering: In early access, Stripe used Fable 5 to perform a code‑base‑wide migration in a 50‑million‑line Ruby project in one day – a task that would normally take a team several months.
- Knowledge work: On Hebbia’s Finance Benchmark for senior‑level reasoning, Fable 5 achieved the highest score, excelling at document‑based reasoning, chart and table interpretation, and complex problem solving.
- Vision: The model can extract precise numbers from scientific figures, rebuild a web‑app’s source code from screenshots alone, and even beat “Pokémon FireRed” using a vision‑only harness.
- Long‑context memory: With a 1 M‑token context window, Fable 5 stays focused across millions of tokens and improves its outputs using its own notes. In the game Slay the Spire, its persistent file‑based memory gave it three times the performance of Opus 4.8.
- Mythos 5 specifics: Internal protein‑design experts saw a ten‑fold acceleration in parts of drug design. In blinded comparisons, scientists preferred its molecular‑biology hypotheses ~80 % of the time. It also ran an autonomous week‑long genomics project, training a custom model on single‑cell data from 138 species that outperformed a recent Science paper model despite being 100× smaller.
How the safeguards work
Because a model this powerful can be misused, Fable 5 ships with new safety classifiers that detect potentially dangerous queries (e.g., cybersecurity, biology/chemistry, and “distillation” topics). When a request is flagged, the response is delegated to Claude Opus 4.8, and the user is informed of the fallback.
These classifiers are deliberately conservative; they trigger in less than 5 % of sessions, meaning that over 95 % of interactions proceed without any fallback, delivering Mythos‑class performance with safety built in. Anthropic’s red‑teaming and external bug‑bounty programs did not uncover a universal jailbreak after more than 1 000 hours of testing.
Mythos 5 removes the cybersecurity safeguards, giving it the strongest offensive capabilities of any current model. It is available only through Project Glasswing in collaboration with the US government.
Use cases for technical teams
- Large‑scale code migration: Long‑horizon reasoning makes big refactors (e.g., 50‑million‑line migrations) feasible in days instead of months.
- Agentic coding pipelines: Fewer interaction turns and higher token efficiency simplify multi‑step autonomous coding agents.
- Finance and analytics: Advanced document, chart, and table reasoning supports senior‑level financial analysis and trading‑strategy development.
- Vision‑to‑code: Converting screenshots or scientific figures directly into source code accelerates front‑end reconstruction and data‑extraction workflows.
- Long‑running research agents: Persistent memory across millions of tokens enables multi‑day autonomous research loops, as demonstrated in genomics and drug‑design experiments.
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 model | Same as Mythos 5 | Same as Fable 5 | Opus 4.8 |
| Availability | Generally available | Limited (Project Glasswing) | Generally available |
| Safety classifiers | Active (cyber, bio/chem, distillation) | Cyber safeguards lifted | Standard Opus safeguards |
| Fallback target | Opus 4.8 | None | None |
| API model ID | claude-fable-5 |
claude-mythos-5 |
(existing Opus ID) |
| Context window | 1 M tokens (default) | 1 M tokens (default) | Per Opus specs |
| Max output | 128 k tokens/request | 128 k tokens/request | Per Opus specs |
| Input price (per 1 M) | $10 | $10 | (Opus pricing) |
| Output price (per 1 M) | $50 | $50 | (Opus pricing) |
| Thinking mode | Adaptive only, always on | Adaptive only, always on | Configurable |
| Data retention | 30 days (Covered Model) | 30 days (Covered Model) | Standard options |
Key takeaways
- Fable 5 and Mythos 5 share the same underlying model; the only difference is the safety classifiers.
- Fable 5 is reported to be state‑of‑the‑art on nearly all capability benchmarks.
- Safety classifiers trigger in under 5 % of sessions, falling back to Opus 4.8 when needed.
- Both models offer a 1 M‑token context window at $10 per million input tokens and $50 per million output tokens.
- Mythos 5 remains limited to Project Glasswing; Fable 5 is generally available across major platforms.
For technical details, see the official announcement and the API documentation. Additional community sentiment analytics can be found at the Mythos Sentiment Observatory.
