Anthropic's Fable 5.1 Costs the Same, Except for the Part Agents Use Most
Claude Fable 5.1 launched on 1 September with unchanged headline pricing at $10 and $50 per million tokens, but cache reads dropped 75 percent to $0.25. Anthropic says that saves about 25 percent on typical work.
Anthropic released Claude Fable 5.1 at the start of last week, along with a restricted twin called Mythos 5.1. The headline prices did not move: $10 per million input tokens and $50 per million output tokens, the same as Fable 5. The change is buried one line further down, and for anyone running agents it is the more important number. Cache reads fell from $1 to $0.25 per million tokens, a 75 percent cut.
Caching is worth a sentence of explanation because it is the sort of plumbing that never makes headlines and quietly decides your bill. When you send a model the same block of text over and over, say a long system prompt, a codebase, or a document you keep asking questions about, the provider can store the processed version and charge you a reduced rate to read it back instead of processing it fresh each time. Anthropic estimates the cut saves around 25 percent on ordinary workloads and up to 45 percent on heavily agentic ones, where a model loops over the same context dozens of times in a single task. Anthropic reports Fable 5.1 scoring 52.6 on Terminal-Bench-Science against 24.7 for Fable 5, though that is a self-reported figure and worth treating as such until independent numbers appear.
Mythos 5.1 is the same underlying model with fewer restrictions, available through vetted-access programmes to cybersecurity and life-sciences organisations that need capabilities the production safeguards normally block. That is now the third lab in a week to ship this exact structure.
Why the cache and not the sticker price. Cutting the headline rate is the loud move, and it invites an immediate price comparison with every rival. Cutting the cache rate is quieter and targets a specific customer: the one running long autonomous jobs, where the same context gets re-read hundreds of times and cache reads dominate the invoice. That is the customer Anthropic most wants, because agentic coding work is sticky and it is where Claude has been strongest. Read as a business move, this is Anthropic making its bill cheapest exactly where it is already winning, without giving away margin on ordinary chat traffic.
What this means for you. If you talk to Claude through the app, none of this touches you. If you pay per token, check whether your setup actually uses prompt caching, because a surprising number do not, and turning it on is usually a small code change for a large saving. If you run coding agents that grind through the same repository repeatedly, this is the release worth reading the pricing page for. And a general habit: when a lab says “same price”, look at the cache line, the batch line and the expiry date on any introductory rate. That is where the real numbers moved this year.
Sources
Google's Gemini 3.8 Flash Keeps the Old Price, and Brings a Cyber Twin
Google DeepMind released Gemini 3.8 Flash on 2 September at $0.75 per million input tokens, unchanged from 3.7 Flash. A separate Cyber variant for vulnerability finding and patching goes only to vetted defenders.
