Claude Fable 5: Anthropic releases its most powerful AI model yet
On June 9, Anthropic released Claude Fable 5 — the first publicly available "Mythos-class" model, with state-of-the-art performance on nearly every capability benchmark tested. It is a real step beyond Opus 4.8, shipped with a new safety architecture. What it can do, what it costs, and what it means for a business — in short, with official sources.
What Fable 5 is
Fable 5 is the public, safeguarded version of Anthropic's most capable internal model (Mythos 5 — the same underlying model, accessible only to vetted research partners). Anthropic describes it as the most capable model it has ever made generally available: state-of-the-art on nearly all capability benchmarks, with exceptional results in software engineering, knowledge work, vision and scientific research. It has a 1-million-token context window and output of up to 128K tokens.
What it can do, in business terms
Months-long projects, compressed into days. The most cited example from the launch: Stripe reported that Fable 5 completed a 50-million-line code migration in a single day — work that would have taken two months manually. For an ordinary company, that translates into migration, integration and data-cleanup projects that used to be too expensive to be worth it — and now aren't.
Senior-level reasoning. Fable 5 posted the highest score on Hebbia's finance benchmark (senior-analyst-level reasoning) and is the top performer on Cognition's FrontierCode coding evaluation. Its strengths: long-horizon reasoning and generalization to new tools — exactly what matters for agents working autonomously for hours at a time.
Precision vision. It extracts exact figures from charts and scientific plots and can rebuild a web app's source code from a screenshot. For workflows built on scanned documents, reports and dashboards, that is a practical difference, not a cosmetic one.
Safety: automatic fallback to Opus 4.8
The architectural novelty: Fable 5 ships with safety classifiers that, on sensitive topics (offensive cybersecurity, dual-use biology/chemistry, model-capability extraction), block the response and hand the conversation over to Claude Opus 4.8 automatically. Anthropic says the safeguards trigger in under 5% of sessions and were deliberately tuned to be cautious — accepting false positives as the price of safety. For ordinary business use, the impact is minimal.
Pricing and availability
$10 per million input tokens and $50 per million output tokens — double the price of Opus 4.8. It is available immediately via the API (model identifier claude-fable-5), in the Claude apps and in Claude Code. On Pro, Max, Team and Enterprise subscriptions it is included at no extra cost until June 22; from June 23 it moves to usage credits.
Why it matters for business
For most day-to-day automations — lead qualification, support, reporting — Opus 4.8 remains the workhorse with the best cost profile. Fable 5 changes the math at the hard end of the spectrum: large migration and integration projects, complex financial analysis, agents that must work autonomously for long stretches without losing the thread. In practice, the threshold of "this can't be automated" just moved up again. Choosing the right model per task — not the most expensive one everywhere — remains the difference between a profitable automation and an expensive one.
Sources: ↗ Official Anthropic announcement · ↗ The Decoder
See what MassAI agents can do →