Claude Fable 5.2

As of November 9, 2026, Claude Fable 5.2 by Anthropic costs $8.00 per 1M input tokens and $40.00 per 1M output tokens, with a 1M-token context window.

Claude Fable 5.2 is Anthropic's top tier as of November 9, 2026, shipping as claude-fable-5-2 alongside the gated Claude Mythos 5.2. The rate card moves down: $8 per million input tokens and $40 output, 20 percent under Fable 5.1 at $10 and $50, with cache reads at $0.20 (still the 0.025x multiplier), cache writes at $10 for five minutes and $16 for an hour, and Batch at $4 and $20. Fast mode is $16 and $80 and generates up to 2.5x faster, and ordinary output is up to 30 percent quicker than 5.1. The shape of the model is unchanged: a 1 million token context window as default and maximum, 128K output, text and vision in, adaptive thinking always on, and the full low through max effort ladder with a default of high. Anthropic's launch table is the first thing to read with care, because the jumps are large: FrontierCode 2.0 at 49.1 percent at xhigh against 32.4 for Fable 5.1, 45.5 for Claude Opus 5.5, 42.8 for GPT-6 Astra, and 41.1 for Gemini 4 Carbon at medium; FrontierSWE v2 at 69.9 percent; Terminal-Bench-Science 0.1 at 74.6; OSWorld 2.1 at 89.3 on the offline partial setting; Humanity's Last Exam with tools at 74.2; the Remote Labor Index at 41.3 percent at high effort against 17.9 for 5.1; and RoboDojo Sim at 64.9 against 27.6. Gemini 4 Carbon still leads AutomationBench, 53.9 to 51.4, and Opus 5.5 holds a higher AA-Briefcase Elo at 1800 against 1785 even as Fable 5.2 leads the rubric score 66.1 to 61.2. Independent measurement has started: Vals AI puts Fable 5.2 first of 36 models on its Vals Index at 71.31 percent as of November 9, 2026. Four changes can break code written for Fable 5.1: the thinking prefix check now returns a 400 on every account rather than only accounts created on or after August 31, 2026, only Fable 5.2 and Mythos 5.2 can read Fable 5.2 thinking blocks, the older computer_20251124 tool returns a 400 everywhere except Amazon Bedrock in favor of computer_toolset_20260801, and an advisor tool executor on 5.2 accepts only a 5.2 advisor. Forced tool choice and disabling thinking both still return 400. Safety classifiers remain, but Anthropic reports roughly 55 percent fewer blocks than Fable 5.1 overall, about 60 percent fewer in cybersecurity and 40 percent fewer in biology; blocked cyber requests are answered by Claude Opus 4.8 and blocked biology requests by Claude Opus 5, which are also the permitted server-side fallback targets. It requires 30-day data retention and is not offered under zero data retention without written authorization. Available on the Claude API, Amazon Bedrock, Claude Platform on AWS, both Google Cloud routes, and Microsoft Foundry, with retirement no sooner than November 9, 2027.