Anthropic launches Claude Fable 5.1 and Mythos 5.1 with cheaper cache pricing, fewer safety-classifier false positives, and incremental benchmark gains

Anthropic released Claude Fable 5.1 and companion model Mythos 5.1. The concrete, checkable changes are pricing and safety-filter tuning: cache read tokens drop from $1.00 to $0.25 per million (Anthropic claims up to 38-45% cheaper for agentic and Claude Code workloads), and false-positive safety-classifier interventions fall by roughly 85% for biology-related queries and 60% for cyber-related ones. A zero-data-retention storage option is promised but not yet shipped, with full ZDR available only to select eligible customers now. OpenRouter lists the model at $10/M input and $50/M output tokens. Benchmark gains over Fable 5/Opus 5 are described as modest.

Anyone running meaningful volume through Claude's API or Claude Code should recalculate costs now, since a 75% cut in cache read pricing is a real, structural saving that will show up in bills regardless of how the model performs otherwise. Lower false-positive rates on safety classifiers matter for developers who have had legitimate biology or security-related queries blocked, potentially reducing friction for research and cybersecurity use cases. Beyond cost and filter tuning, this is an incremental point release, not a capability leap, so it should not change build-vs-buy or vendor-lock decisions on its own.

The "world's most powerful models" framing (applied to both Fable 5.1 and rival GPT-6 Astra) is marketing packaging, not a substantiated claim, and sits alongside vague qualitative praise about writing and agentic tool use with no benchmark table cited in this account. The source is a LessWrong commentary piece relaying quoted Anthropic staff statements, not the official launch blog, model card, or benchmark release, so the specific percentages should be treated as one step removed from the primary record until verified against Anthropic's own documentation. The zero-data-retention feature is described as forthcoming, not live, so treating it as a current capability would be premature.

1. 01Check Anthropic's official model card or launch blog to confirm the $1 to $0.25 cache price cut and the 85%/60% false-positive reduction figures against primary documentation.
2. 02Track actual enterprise adoption or usage volume shifts (e.g. via OpenRouter usage stats or Anthropic customer disclosures) over the next one to two months to see if the price cut drives measurable uptake.
3. 03Watch for the general availability date and terms of the zero-data-retention option, since it is currently limited to select eligible customers rather than shipped broadly.