Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1, two versions of the same model with different safeguards, with lower cache-read pricing, changes to how thinking blocks work in the API, new controls for enterprise data, updated safeguards, and content provenance features - Fable 5.1 is generally available and Mythos 5.1 is limited to approved users through trusted-access programs for cybersecurity and life sciences, both have a 1M-token context window, up to 128k output tokens and always-on adaptive thinking, and Anthropic recommends Fable 5.1 mainly for demanding reasoning and long-running agentic work where Opus 5 at higher effort still falls short - Anthropic reports higher results than Fable 5 across many coding, research, knowledge-work and computer-use benchmarks, Mythos 5.1 produced protein binders with binding affinities 10 times higher than the best designs in the referenced competitions on three targets, Fable 5.1 created a higher-resolution elevation map covering one-third of Venus, and Mythos 5.1 sped up seven open-source biology models by up to 2.5 times - Input and output API prices remain $10/M and $50/M tokens, but cache reads now cost $0.25/M, 75% less than before, Anthropic estimates around 25% lower cost for typical Fable workloads and up to around 45% for complex coding and highly agentic work, and Enterprise Frontier Safeguards will let eligible customers keep activity data in their own cloud account, use customer-managed encryption keys, and have automated safety monitoring send misuse flags directly to their own teams with no Anthropic human review required - The main API breaking changes are that forced "tool_choice" no longer works, earlier Claude models cannot read Fable 5.1 thinking blocks, and changing the system prompt, tools or earlier messages invalidates later thinking blocks - this check is active by default for new API accounts created on or after August 31, 2026, developers can choose to drop affected blocks instead of failing the request, and Anthropic says the change closes off a common, publicly documented technique used to extract Claude's thinking for model distillation - Fable 5.1 adds per-message effort, turn-scoped system messages and readable progress updates between tool calls, text generated by Fable 5.1 and Mythos 5.1 carries Anthropic's statistical watermark, and supported generated images and videos can include signed C2PA Content Credentials when retrieved through the Files API I've been using Fable 5.1 since yesterday, and so far it looks like a smart model, but for me it is still unusable for larger tasks, even on Max 20x (the highest plan) - one review task with ~40 subagents, limited to high reasoning effort, used my full 5h limit and about 40% of my weekly usage limit in ~30min (