Fable 5 First Look: What Anthropic's New Flagship Means

Anthropic's model lineup has had a stable shape for years: Haiku for speed, Sonnet for balance, Opus at the top. Fable 5 breaks that shape — it's a new tier above Opus, priced at $10/$50 per million tokens, double Claude Opus 4.8.

The headline facts

• Model ID: claude-fable-5
• Context: 1M tokens, standard pricing — no long-context premium
• Output: up to 128K tokens
• Thinking: adaptive only; sampling parameters are gone entirely

What actually feels different

The spec sheet undersells it. In day-to-day use, three things stand out:

It stays coherent longer. On multi-hour agentic tasks — big refactors, research sessions with dozens of tool calls — Fable 5 holds the thread where earlier models would drift or need re-anchoring.

It knows when to think. Adaptive thinking means simple questions come back fast, while genuinely hard problems get visibly deeper reasoning. You stop tuning budgets and start trusting the model's own judgment.

The ceiling is real. On the hardest prompts we throw at models — multi-constraint reasoning, subtle debugging, expert-level writing — Fable 5 succeeds where Opus 4.8 plateaus. For everyday tasks the gap is smaller, which is exactly why routing by difficulty matters.

Should you switch?

If your workload already pushes Opus 4.8 to its limits, yes — the API surface is identical, so it's a model-ID swap. If not, Opus 4.8 at half the price remains a great default. We've written a full comparison if you want the details.

Or skip the reading and chat with Fable 5 directly in our playground — free credits on sign-up.