Overview
- Anthropic announced Tuesday that it is launching Claude Fable 5.1 for broad use and a restricted Mythos 5.1 for vetted cybersecurity and life-sciences partners, with both sharing the same core model but different safety guardrails.
- Fable 5.1 supports very large contexts and long outputs, with one million tokens in context and up to 128,000 tokens per response, and Anthropic reports big gains on multi-step science benchmarks such as Terminal-Bench-Science (52.6% versus 24.7% for Fable 5).
- The company cut cache-read prices by 75% to $0.25 per million tokens while keeping base rates at $10 per million input tokens and $50 per million output tokens, a move Anthropic says will lower typical workload costs by roughly 25% and by up to about 45% for heavily agentic workloads that repeatedly re-read context.
- Anthropic is rolling out Enterprise Frontier Safeguards so customers can keep model inputs on customer-controlled cloud storage with zero data retention for eligible users until EFS ships, and it is opening an invisible watermark detection API in private preview for approved regulators and researchers.
- Anthropic tightened safety classifiers to reduce false positives — about 60% fewer cyber interventions and about 85% fewer benign biology blocks — while keeping exploit generation and other high-risk actions blocked, and it plans phased EFS and oversight tool rollouts this fall under continuing regulatory and security scrutiny.