ক্লড ফেবল 5

Anthropic's most capable widely released model, built for the most demanding reasoning and long-horizon agentic work. Its capabilities exceed anything Anthropic had previously made generally available, and it is state of the art on nearly all tested benchmarks, with exceptional results in software engineering, knowledge work, vision and scientific research. It can work autonomously for longer than any previous Claude, and it always reasons -- thinking cannot be switched off.

Fable 5 and Claude Mythos 5 are the same underlying model launched twice. Mythos 5 goes to cyber defenders and critical infrastructure providers with safeguards lifted; Fable 5 is that model made safe for general release, and it is the one served here. Because it always reasons, there is no non-thinking mode to fall back to - every call spends reasoning tokens, and the effort ladder from low to max sets how many rather than whether. Anthropic characterises the advantage as one that widens with the job: the harder and more drawn-out the work, the further Fable 5 pulls ahead of the rest of its range.

The safeguards are the defining fact about this model and worth understanding before you rely on it. Classifiers watch for requests about cybersecurity, biology and chemistry, or attempts to distil the model, and hand those to Claude Opus 4.8 instead, telling the user when it happens. Anthropic tuned them deliberately conservatively, so benign requests are sometimes caught; the lab's early data shows fallback in under 5% of sessions, and says that for the other 95% Fable 5 performs effectively as Mythos 5 does. That reads across into the published benchmark table: on the starred rows, which include Humanity's Last Exam, Terminal-Bench 2.1 and the biology and cybersecurity evaluations, Fable 5 scores closer to Opus 4.8 than the headline figure because of those fallbacks.

On software engineering, Anthropic reports the highest score among frontier models on Cognition's FrontierCode evaluation even at medium effort, and cites Stripe compressing a codebase-wide migration across 50 million lines of Ruby into a day of work that would otherwise have taken a team over two months. Vision is where the lab claims a clean state of the art: extracting precise numbers from detailed scientific figures, rebuilding a web app's source code from screenshots alone, and finishing Pokemon FireRed with a vision-only harness where earlier Claude models needed a scaffold of navigation aids and game-state tools.

Long-horizon behaviour is the other theme. Anthropic says Fable 5 holds focus across millions of tokens and improves its own output from notes it wrote earlier; giving it persistent file-based memory in the deck-building game Slay the Spire helped it three times more than the same memory helped Opus 4.8, and it reached the final act three times as often. On knowledge work the lab reports the highest score of any model on Hebbia's Finance Benchmark, with the gains concentrated in document-based reasoning and chart and table interpretation.

Two operational notes come with the capability. Traffic on Mythos-class models carries mandatory 30-day retention on first- and third-party surfaces, which Anthropic says it uses for safety work and jailbreak detection and not for training. And on alignment the lab reports the model's rate of misaligned behaviour as low and similar to Opus 4.8's, measured on Mythos 5 and inherited by Fable 5 because they are the same weights.