Claude Fable 5.2

Anthropic makes Claude, one of the AI systems that work on Spellwright's Intelligence dossiers.

Closed

Anthropic released Claude Fable 5.2 on November 9, 2026. Its Claude API ID claude-fable-5-2 names a pinned snapshot under Anthropic's versioning convention. Anthropic positions it for demanding reasoning and long-horizon agentic work and offers it to all users with cybersecurity and biology safeguards.

Benchmark scores

Humanity's Last Exam · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.74.2%#1 of 5 results
AutomationBench · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.51.4%#1 of 2 results
Terminal-Bench-Science 0.1 · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.74.6%#2 of 2 results
AA-Briefcase v1.1 · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.66.1%#1 of 1 results
AA-Briefcase (Elo) v1.1 · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.1785 points#1 of 1 results
BenchCAD 2.1 · Reported by the model's maker, Nov 9, 2026Setting: Max effort, as stated in the chart description. Numeric budgets not stated.83.5 points#1 of 1 results
Chartography · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.73.4%#1 of 2 results
Chartography · Reported by the model's maker, Nov 9, 2026Setting: Medium effort, as stated in the chart description. Numeric budgets not stated.66.4%#2 of 2 results
FrontierCode 2.0 · Reported by the model's maker, Nov 9, 2026Setting: Xhigh effort, as labeled in the table row. Numeric budgets not stated.49.1%#1 of 2 results
FrontierCode 2.0 · Reported by the model's maker, Nov 9, 2026Setting: High effort, as stated in the chart description. Numeric budgets not stated.45.3%#2 of 2 results
FrontierSWE v2 · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.69.9%#1 of 1 results
OSWorld 2.1 · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.89.3%#1 of 1 results
Remote Labor Index · Reported by the model's maker, Nov 9, 2026Setting: High effort, as labeled in the table row. Numeric budgets not stated.41.3%#1 of 1 results
RoboDojo Sim · Reported by the model's maker, Nov 9, 2026Setting: Max effort; Anthropic says this is the highest-available level and that it ran the evaluation internally.64.9%#1 of 1 results
WANDR · Reported by the model's maker, Nov 9, 2026Setting: Highest-available effort, per the table footnote; the effort level is not named for this row. Numeric reasoning and token budgets are not stated.82.1%#1 of 1 results

Access

Anthropic's overview lists the Claude API, Amazon Bedrock, Google Cloud (Google managed and Anthropic managed), Microsoft Foundry and Claude Platform on AWS; the announcement says it is available on all platforms from launch day. Anthropic says safety classifiers block high-risk cyber work and most research-biology tasks, routing those queries to Opus 4.8 or Opus 5 instead. It requires 30-day data retention unless Anthropic expressly authorizes zero data retention. No API call, account entitlement or regional deployment was tested.