Claude Fable 5 Is the Most Powerful Model Anthropic Has Released to the Public. Here Is What to Hand It First.

What You Will Learn

• What "Mythos-class" actually means and how Fable 5 differs from Opus 4.8
• Why the one-million-token window changes which jobs are worth handing it
• The four tasks that pay back fastest in the first week
• How the automatic safety routing works and when it kicks in
• What free access until June 22 means and how to test before it ends

For months, Anthropic's most capable system, the Mythos family, was something most people only read about. It scanned codebases for security firms and assisted a handful of approved researchers. On June 9 that changed. Claude Fable 5 is the first Mythos-class model released for general use, and if you pay for Claude Pro, Max, Team, or Enterprise, it is already in your model picker.

The number that matters most is one million. Fable 5 can hold up to a million tokens of context at once, roughly the length of a small data room, and it can work autonomously on a long task for longer than any Claude before it. Anthropic positions it above the Opus line for difficult reasoning, long-horizon agent work, coding, and large-document analysis. It costs ten dollars per million input tokens and fifty per million output on the API, but the more relevant fact for most readers is that it carries no extra cost inside existing plans until June 22.

For an executive, the headline is not the benchmark. It is that the kind of read that used to require a junior analyst and an afternoon, going through a pile of long documents to surface what matters, now fits inside a single supervised session. The question is which jobs are worth it and how to keep the run honest. That is the rest of this guide.

This is a PromptHacker Premium deep dive.

The four jobs to hand Fable 5 first, how the safety routing behaves, and the way to test it before free access ends are below.

What "Mythos-Class" Means, in Plain Terms

Fable 5 shares the underlying model family with Claude Mythos 5, Anthropic's research-grade system, but adds stricter safeguards so it is safe for broad release. Think of it as the same engine with guardrails bolted on. Compared with Opus 4.8, the model most readers used last week, Fable 5 is built for two things in particular: holding far more material in working memory at once, and staying coherent across a longer sequence of steps without losing the original goal.

Plain-English takeaway: you do not need to chop a long document into pieces anymore. Paste the whole thing and ask one question across all of it.

How the Safety Routing Works

Fable 5 ships with conservative safeguards. When a request touches a high-risk domain, specifically cybersecurity, biology and chemistry, or model distillation, the system quietly hands that question to Opus 4.8 instead of answering with the Mythos-class engine. Anthropic expects this to trigger in under five percent of sessions. For ordinary business work, finance, marketing, operations, contracts, you will never notice it. It matters only if your work sits near those sensitive areas, in which case some answers will come from the older model.

The Four Jobs to Hand Fable 5 First

Pick work that is long, document-heavy, and currently eating real hours. The million-token window is wasted on a one-paragraph question, so save Fable 5 for the jobs below.

1. The full data room read

Considering an acquisition, a partnership, or a vendor switch? Drop the entire folder of agreements, financials, and reports in at once and ask for the risks, the obligations, and the open questions, each with a citation to the source line. This is the Pro Tip this week, with the exact prompt.

2. The long contract or policy set

A master services agreement plus its amendments and exhibits can run a hundred pages across several files. Fable 5 reads them together and reconciles them, flagging where an amendment changes a clause in the base contract, the kind of cross-reference that is easy to miss by hand.

3. The quarter of reports, in one pass

Paste three months of weekly updates, board notes, and metrics, then ask Fable 5 to write the narrative for the quarter: what trended, what stalled, and the three decisions the data points to. Because it holds all of it at once, the story is consistent rather than stitched together from separate chats.

4. The long-horizon project plan

Hand it a goal, the constraints, and the relevant background documents, and let it produce a phased plan with dependencies and a first draft of each phase. The longer-run improvement means it can carry the plan through to a usable draft instead of stopping at an outline.

Test It Before Free Access Ends

After June 22, Fable 5 use inside Pro, Max, Team, and Enterprise moves to a credit pool, so the smart move is to run a real job this week and decide whether it earns a place in your routine. Use a task you already know the answer to, a contract you have read or a quarter you lived through, so you can judge the quality against ground truth rather than trusting it blind.

Action Steps Summary

• Select Fable 5 in your model picker. It is live in Claude web, desktop, and mobile for Pro, Max, Team, and Enterprise.
• Pick one long, document-heavy job. A data room, a contract set, or a quarter of reports, not a quick question.
• Paste everything at once. The million-token window means you no longer split documents into chunks.
• Ask for citations and confidence labels. Have it point to the exact source line and flag where it is unsure.
• Judge it against something you know. Run a task with a known answer before June 22, then decide if it earns credits.