Claude Fable 5.2 cuts price 20% to $8 in, $40 out

Anthropic released Claude Fable 5.2 and Mythos 5.2 today, and the Fable 5.2 price is 20% lower than Fable 5.1: $8 per million input tokens and $40 per million output, down from $10 and $50. Cache reads fall to $0.20. Anthropic also says Fable 5.2 blocks about 55% fewer requests than 5.1. Cheaper and less jumpy is a good combination. It is still the most expensive tier in the Claude lineup, so the question for an operator is not "upgrade?" It is "which jobs earn this model?"

What actually happened

From Anthropic's Fable and Mythos 5.2 announcement, the pricing docs and the models overview:

• The price. Fable 5.2 and Mythos 5.2: $8 input, $40 output, $0.20 cache reads, $10 for five-minute cache writes. Batch is half that: $4 in, $20 out.
• The model ID. claude-fable-5-2 on the Claude API, Google Cloud and Microsoft Foundry; anthropic.claude-fable-5-2 on Amazon Bedrock. 1M-token context, 128K max output, standard pricing across the full window.
• Fewer blocks. Anthropic says about 55% fewer safeguard blocks than 5.1: about 60% fewer in cybersecurity, about 40% fewer in biology. When a request is still blocked, it falls back to an Opus model instead of failing.
• Mythos stays gated. Mythos 5.2 is the same underlying model without those safeguards. It is limited to organizations in Anthropic's verification programs.
• The benchmarks. Anthropic reports 69.9% on FrontierSWE v2 (5.1: 56.3%) and 51.4% on AutomationBench (5.1: 31.4%). Not a clean sweep: Anthropic's own table shows Gemini 4 Carbon at 53.9% on AutomationBench. These are vendor numbers.
• 5.1 is not going away yet. The deprecations page lists Fable 5.1 as active, retiring not sooner than September 1, 2027.

TechCrunch covers the safety backdrop, including Anthropic's statement that it limited some AI-research capabilities in training.

Why Fable 5.2 pricing matters for your business

A 20% cut on the top tier changes the routing math, not the routing rule. Most small-business AI work is bounded: tag this ticket, pull fields from this invoice. That does not need a frontier model at $40 per million output tokens. The jobs that do are the long, multi-step ones where a wrong turn costs an hour of cleanup: a codebase change, a messy data migration, a contract review with real money on it.

So rerun the numbers on your actual logs. Take last month's Fable traffic, price it at $8/$40, then price the same traffic with the bounded work moved to a cheaper model. If you run long agent sessions, the $0.20 cache-read rate matters more than the headline. Measure your cache-hit rate first. And swap model IDs in staging, not production. One more check: the deprecations page says API version 2026-10-21 rejects temperature, top_p and top_k with a 400 error. If your code still sends them, find out before your customers do.

Sources: Anthropic: Claude Fable and Mythos 5.2, Claude API pricing, Claude model deprecations, TechCrunch.