Claude Fable 5 is Anthropic's most capable publicly available model — the first from its powerful "Mythos" class anyone can use. This guide is the practical how-to: how to get access, how to call it, when to pick it over Claude Opus 4.8, and how to keep the bill (and the guardrails) under control. For the full verdict and benchmarks, see our Claude Fable 5 review.
Getting access
Fable 5 is available three ways: through the Claude API (model id claude-fable-5), on the Claude apps via Pro, Max, Team and Enterprise plans, and inside GitHub Copilot. There is a launch window worth noting: from June 9 through June 22, 2026, Fable 5 is bundled at no extra cost on Pro, Max, Team and seat-based Enterprise plans. After June 22 it draws on usage credits, so plan any heavy experimentation before the cliff.
Calling Fable 5 from the API
If you already use the Claude Messages API, switching is a one-line change — just set the model to claude-fable-5:
curl https://api.anthropic.com/v1/messages -H "x-api-key: $ANTHROPIC_API_KEY" -H "anthropic-version: 2023-06-01" -H "content-type: application/json" -d '{"model":"claude-fable-5","max_tokens":1024,"messages":[{"role":"user","content":"Refactor this function and explain the change."}]}'Everything else (tools, streaming, system prompts) works exactly as it does for other Claude models. Check the official API docs for the current version header and the full parameter list.
Fable 5 or Opus 4.8? A quick rule
Fable is the better model, but it is not always the right one. A simple heuristic: reach for Fable 5 on your hardest work — complex agentic coding, long multi-step tasks, and vision-heavy jobs where its accuracy advantage pays off. Default to Opus 4.8 for high-volume, cost-sensitive, or simpler workloads, where it delivers most of the quality at half the price. On the toughest coding tasks Fable's accuracy keeps climbing the more reasoning budget you give it; on routine ones, that extra spend is wasted.
Managing cost
Fable 5 lists at $10 per million input tokens and $50 per million output — roughly double Opus 4.8 and the priciest of the major frontier models. Output is where the bill grows, so the biggest levers are simple: ask for concise outputs, avoid re-sending huge contexts on every call, and use a lower reasoning effort for easy tasks. Reserve maximum effort for the problems that actually need it.
Working with the guardrails
Fable runs three classifier-based safeguards — for offensive cyber, biology and chemistry, and model distillation. When a request trips one, Fable declines and routes it to Claude Opus 4.8 instead. Anthropic says over 95% of sessions never hit a guardrail, so most users will not notice. Two things to know: if you do legitimate security or red-team work, expect blocks (the guardrails are designed to stop exactly that), and a mandatory 30-day data-retention window applies to all Mythos-class traffic — even on zero-retention plans — so factor that into any sensitive workload.
Tips to get the most out of it
Lean on Fable's strengths: give it long, multi-step tasks it can work through autonomously rather than micromanaging each step; feed it screenshots, diagrams and PDFs, since it is state-of-the-art at vision; and tune the reasoning effort to the difficulty of the task to balance quality against cost. For anything that touches blocked categories, send it to Opus 4.8 directly rather than fighting the classifier.
That is the practical picture. For the numbers behind it, see our review; for the launch details and what Fable means for the wider market, read our launch coverage.