Claude Fable: Anthropic's Top Reasoning Line
Current status
The current flagship of the line is Claude Fable 5.1 (released September 2026). Pricing, context window and benchmarks live there — this page describes the line as a whole and doesn’t age with every new version.
What Claude Fable stands for
Fable is Anthropic’s top reasoning line — the tier that in 2026 reaches beyond the classic three-way split of Haiku, Sonnet and Opus. Where Opus is the regular pro line, Fable sits one floor higher: even more reasoning depth, at the cost of being pricier and slower. It’s not an everyday model but one for the tasks where the last few percent of reliability make the difference.
One thing that often confuses on first contact: Fable and Mythos are the same underlying model. The difference is only in the safeguards. Fable runs with stricter safeguards and is broadly available; Mythos runs with more permissive safeguards and is accessible only through Anthropic’s trusted-access programs for vetted cybersecurity and life-science organizations. So for the vast majority of teams, “use this line” means: Fable.
- Coding 5 / 5 · Feld-Spitze
- Reasoning 5 / 5 · Feld-Spitze
- Text 5 / 5 · Feld-Spitze
- Vision 4.5 / 5 · Gemini 3.1 Pro
- Speed 3 / 5 · Claude Haiku 4.5 u. a.
- Kosten-Eff. 2.5 / 5 · GPT Luna u. a.
Eignung 0–5 · redaktionelle Einordnung, kein Benchmark · gestrichelt = Feld-Bestwert je Achse
The profile shows the frontier signature: top marks on coding, reasoning and text, strong on vision — and, as the price for that, the lowest scores in the field on speed and cost efficiency. Fable buys you maximum depth, not throughput.
Where it fits
Fable is worth it where reasoning depth trumps everything and budget is the second question:
- Long-running coding agents — multi-hour runs in large codebases where an abort is expensive and reliability across the chain counts.
- Complex multi-step analysis — nested research, proofs, argumentation where one logical slip voids the whole result.
- Large document sets — contracts, studies, codebases in combination, where maximum consistency across very long context matters.
Gotcha: thinking can't be turned off
On Fable, thinking is adaptive and always on — it can’t be disabled on any effort level or via the API; the effort setting defaults to “high”. The remaining cost lever is to turn the effort level down rather than disable thinking. If you migrate from a model without forced thinking, recalculate the token cost.
How to choose: Fable or Opus?
The real decision with Fable isn’t “which Claude line” but “is Opus enough, or do I need the floor above it?”. The positioning map shows how close the frontier models sit together — and how far they are from the volume zone:
Redaktionelle Einordnung, kein Benchmark
As a rule of thumb:
- Do the last few percent of reasoning reliability make the difference, and the budget carries it? Then Fable.
- Do you need solid frontier quality at regular pro pricing? Then Opus — the broadly available default for demanding projects.
- Does volume or latency matter? Then no frontier line gets you there — reach for Sonnet or Haiku.
A favourable lever on Fable 5.1: the cost for cached input tokens dropped by roughly 75 percent versus Fable 5. For agent workflows with a lot of reused context, that shifts the math noticeably.
Read on
- Claude Fable 5.1 — the current flagship with pricing and benchmarks
- Claude Fable 5 — the previous version of the line
- Claude Mythos 5.1 — the same model with more permissive safeguards (trusted access)
- LLM model families at a glance — Fable next to GPT, Gemini and co.
- AI pricing explained — how the Fable price and caching pencil out
FAQ
- Fable is Anthropic's top reasoning line — even more capable than Opus, but also pricier and slower. Sonnet and Opus remain the more economical lines for everyday and balanced frontier work respectively.
- No. On Fable, thinking can't be disabled on any effort level or via the API. The remaining cost lever is to turn the effort level down rather than disable thinking.
- Both names refer to the same underlying model. Fable runs with stricter safeguards and is broadly available, Mythos with more permissive safeguards and only through Anthropic's trusted-access programs for vetted cybersecurity and life-science organizations.
- Fable 5.1 adds image input and, above all, cuts the cost for cached input tokens by roughly 75 percent — a noticeable advantage for agent workflows with a lot of reused context.
- For teams with long-running coding agents, complex multi-step analyses or large document sets where maximum reasoning depth makes the difference. For high-throughput tasks on a tight budget, Sonnet and Haiku remain the better pick.