Claude Fable 5.1
Language Available Full comparison ↗ anthropic/claude-fable-5-1 · by Anthropic
· decoder-only-transformer
Pricing — 1 offering(s)
Input tokens
- $10.00 / 1M tokens (input)
Current
2026-09-23 → present
Same as Fable 5. The models-overview table shows the same figure. First-time entry, so effective_from is this…
Same as Fable 5. The models-overview table shows the same figure. First-time entry, so effective_from is this verification date; the model launched in early September 2026.
Output tokens
- $50.00 / 1M tokens (output)
Current
2026-09-23 → present
Same as Fable 5. The models-overview table shows the same figure. First-time entry, so effective_from is this…
Same as Fable 5. The models-overview table shows the same figure. First-time entry, so effective_from is this verification date; the model launched in early September 2026.
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Claude Fable 5.1 fits into a cost-aware routing setup
See how →About Claude Fable 5.1
Claude Fable 5.1 is Anthropic's top generally available model, announced in September 2026 alongside Claude Mythos 5.1, a limited-availability model not tracked here. It succeeds Claude Fable 5, which Anthropic now lists as a legacy model. It's a decoder-only transformer with adaptive thinking always on, steered by a five-level `effort` parameter that defaults to high on the Claude API.
List pricing is unchanged from Fable 5 at $10 / $50 per million input / output tokens, but cache reads dropped 75% to $0.25 per million tokens. Anthropic estimates that makes typical workloads about 25% cheaper than on Fable 5, and highly agentic ones up to about 45% cheaper. It has a 1M-token context window and 128K maximum output, and is available on the Claude API, Amazon Web Services, Google Cloud and Microsoft Azure.
Capability profile
How Claude Fable 5.1 rates across core capability dimensions, with the task-level evidence behind each rating.
Anthropic positions Fable 5.1 'for demanding reasoning and long-horizon agentic work' (models-overview) and reports higher scores than Fable 5 across its launch charts (HLE with and without tools, Terminal-Bench-Science 0.1). Those are vendor-reported and not recorded as structured scores here; see limitations.
Launch post: among 'the world's most advanced models for coding and knowledge work'. The launch table reports Terminal-Bench 4.0 at 55.8%, a version the coding vertical doesn't track, and doesn't state effort settings. Vendor-reported; no companion modelglass-coding entry.
Launch table: OSWorld 2.0 77.9% partial / 41.7% strict (vendor-reported, settings not stated). Positioned for long-horizon agentic work.
1M-token context window and 128K max output (models-overview table).
Comparative latency is listed as 'Slower' in Anthropic's models-overview table, the slowest of the current lineup (Opus 5.5 'Moderate', Sonnet 5 'Fast'). Fast mode isn't offered for Fable 5.1.
$10/$50 per MTok, 2.5x Opus 5.5's $4/$20. Cache hits at $0.25/MTok (0.025x base input) make heavily cached, agentic workloads much cheaper than Fable 5: Anthropic estimates ~25% lower cost for typical workloads and up to ~45% for highly agentic ones (vendor claim).
Ratings are estimated — limited independent data is available for this model.
Operator guidance
Anthropic's top generally available model. Its models overview recommends starting with Claude Opus 5.5 for most workloads and moving to Fable 5.1 only when evals at higher Opus 5.5 effort still fall short, so route here for the hardest reasoning and long-horizon agentic work, not by default. It's the slowest current Claude model and costs 2.5x Opus 5.5 per token, though the cheap cache reads narrow the gap on heavily cached agentic loops. Replaces Claude Fable 5, now legacy.
Use cases
- Demanding reasoning and long-horizon agentic work (Anthropic's own stated positioning)
- Workloads where evals at higher Opus 5.5 effort still fall short (Anthropic's models-overview guidance)
- Long, cache-heavy agentic sessions, where the 0.025x cache-read price matters most
Limitations
- No structured benchmark scores are recorded. The launch post's table gives versions (Terminal-Bench 4.0, OSWorld 2.0, HLE, CursorBench 3.2.0, and GDPval-AA v2 1,853) but doesn't state the effort or settings behind each figure, and every figure is vendor-reported. The Fable 5.1 / Mythos 5.1 system card is the place to source settings-disclosed scores for a follow-up.
- The launch table's GDPval-AA figure (1,853) is on the superseded v2 scale. The current GDPval-AA v2.1 score (1,735 at max effort, Artificial Analysis's independent leaderboard) is in the companion modelglass-gdpval entry (SCO-632).
- Exact launch day isn't stated on the launch post (September 2026). Our news feed recorded it on 2026-09-02.