Claude Fable 5.1

Proprietary1M context

Claude Fable 5.1 is a proprietary model from Anthropic with a 1M-token context window. 4 providers list it; as of 25 Oct 2026, the cheapest listings (Anthropic and 2 others) charge $10.00 per million input tokens and $50.00 per million output tokens.

$10.00
Lowest input, per 1M
Same as Anthropic's own API
$50.00
Lowest output, per 1M
$0.25
Cached input, Anthropic's API
4
Providers listed
Joint 13th
GPQA Diamond rank
of 221 models with a score

Providers serving Claude Fable 5.1

Cheapest first, by the blended rate of three input tokens to one output token. Prices in US dollars per million tokens, checked 25 Oct 2026; the source link opens where the price is published. All 4 list tool calling and JSON mode; none lists an OpenAI-compatible API.

What each provider charges

Blended price per 1M tokens (3 input : 1 output); bars start at $0. Prices checked 25 Oct 2026.

What this shows: the highest listing is 10% above the lowest: $22.00 (Amazon Bedrock) against $20.00 (Anthropic and 2 others).

  • Anthropic and 2 others: $20.00 per 1M tokens, blended
  • Amazon Bedrock: $22.00 per 1M tokens, blended
Providers serving Claude Fable 5.1, cheapest first by blended rate
ProviderInputOutputContextModel IDSource
AnthropicMaker's API$10.00$50.001Mclaude-fable-5-1
Google Vertex$10.00$50.001Mclaude-fable-5-1@default
Microsoft Azure$10.00$50.001Mclaude-fable-5-1
Amazon Bedrock$11.00$55.001Manthropic.claude-fable-5-1

How each provider handles your prompts

For each provider's paid API, from its own pages, checked 26 Oct 2026. Where a promise lives matters: a signed agreement binds the provider, while a docs page can change without notice. Not legal advice.

How each provider serving Claude Fable 5.1 handles your prompts
ProviderTrains on your promptsKeeps promptsZero retentionData processing agreement
Anthropic

Flagged content can be kept up to 2 years; Covered Models (Claude Fable 5/5.1, Mythos 5/5.1) require 30-day retention and are excluded from ZDR.

Flagged content can be kept up to 2 years; Covered Models (Claude Fable 5/5.1, Mythos 5/5.1) require 30-day retention and are excluded from ZDR.MoreLess
No
Deleted within 30 days
On request
Public
Google Vertex
Claude Fable/Mythos keep all traffic 30 days and require sharing it with Anthropic; Google Search grounding keeps queries 3 days; the 24-hour cache can be switched off; the Interactions API stores prompts unless store=false.MoreLess
No
Not stored beyond the request, except a 24-hour cache (on by default) and abuse-flagged prompts (up to 90 days)
Enterprise plans only
Public
Microsoft Azure
Approved 'modified abuse monitoring' turns off the storage and human review; stateful features (Responses API, stored completions, Assistants threads) keep history as you configure them.MoreLess
No
Prompts and outputs flagged by abuse monitoring may be stored for human review; no retention period stated
Enterprise plans only
Public
Amazon Bedrock
OpenAI models (flagged traffic) and Claude Fable 5/5.1 (all traffic, with possible AWS human review) keep data up to 30 days; the Responses API stores responses unless store=false.MoreLess
No
Not stored for most models; up to 30 days for abuse detection on models AWS lists
Yes, by default
Public

Claude Fable 5.1 benchmark scores

Compare every model in Bench rankings

Published scores only, each linked to its source. Ticks are every other model's best published score.

Each score is the model's best published result across its settings, named beside it. A setting above the default can use more tokens per task, which a per-token price does not show.

Show as table
Benchmark scores for Claude Fable 5.1, with the setting each was measured at
BenchmarkScoreSetting
GPQA Diamond
PhD-level scientific reasoning, pass@1, no tools
93.7%, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on
APEX-Agents
Agentic office work, APEX-Agents 1.1, Pass@1 in the maintainers' harness
68.6%, source: APEX-Agents leaderboard (Mercor), max effortmax effort
Humanity's Last Exam (text-only)
Reasoning & knowledge, no tools
59.1%, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on
τ³-Banking
Agentic tool use with knowledge retrieval (banking)
47.2%, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on
Terminal-Bench 2.1 (Terminus 2)
Agentic coding & terminal use
91.4%, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on
CritPt v2
Physics reasoning
72%, source: CritPt v2 leaderboardDefault
SciCode (with background)
Scientific coding
63.1%, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on
GDPval-AA v2.1 (Elo)
Real-world work tasks; a rating against other models, not a percentage
1755, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on
LCR
Long-context reasoning, AA-LCR v1.1
85.3%, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on
Omni Accuracy
Omniscience knowledge accuracy
67.2%, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on
Non-Hallucination
1 − hallucination rate
34.4%, source: Artificial Analysis, low effort, production safeguards onlow effort, production safeguards on
ITBench
Kubernetes incident root-cause analysis, Artificial Analysis's ITBench-AA
49.5%, source: Artificial Analysis, max effort, production safeguards onmax effort, production safeguards on

More from Anthropic

QAI lists these prices; it does not serve or resell these models. Prices are each provider's standard on-demand list price in US dollars per million tokens (default region, standard tier, peak hours, base context tier), checked on the date shown against the provider's own published prices or, where it has no public price page, its listing on OpenRouter. Most benchmark scores come from Artificial Analysis; Epoch AI (GPQA Diamond), MathArena (AIME 2026) and the APEX-Agents and CritPt leaderboards add the rest, and where none of them has a model, or none ran it at full capability, a figure its maker published may stand in (marked v); in some benchmarks it was run the maker's own way, which its note names. Each benchmark follows one test setup, the one used by the source that covers the most models, and leaves results from other setups out. Each score is the best its source published across the model's settings, names any non-default setting, and links to its source.

An AIME 2026 score noted “released after the exam” comes from a model released after the exam, so the problems may have been in its training data. GDPval-AA is a rating against other models (Elo), not a percentage. Data as of 26 Oct 2026.

Sources and licences