Claude Fable/Mythos 5.1: The SOTA Model That Changes Agent Economics
Anthropic just shipped new frontier models that dominate every benchmark while cutting cache prices 75% — but at a cost. This is the week agent economics reshuffled.
The Launch
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 as its new flagship models for coding and knowledge work. The announcement drew over 12 million views. Fable 5.1 leads across every evaluation, including 52.6% on Terminal-Bench-Science 0.1 and 55.8% on Terminal-Bench 4.0.
"We're introducing Claude Fable 5.1 and Claude Mythos 5.1. They're the world's most advanced models for coding and knowledge work." — @claudeai
The Pricing Twist
List pricing stayed the same: $10/$50/$12.5 per million tokens. But cache read prices dropped 75% to $0.25/MTok. The catch: 1.7x more output tokens per task, resulting in a net 20% per-task cost increase. The cache discount is real, but the output tax is real too.
What Was Crossed
Fable 5.1 is positioned explicitly around "complex, multi-step work that runs on its own." Community research suggests Fable and Mythos 5.1 may share the same underlying weights with different safety and routing — one model, two personalities.
The Wave
Grok 4.7 and Gemini Flash 3.8 are also on the way. And World Labs Astra — Fei Fei Li and Justin Johnson's world model — launched what the Latent.Space newsletter calls "by far the most impressive world model launch we've ever seen."
For PromptEngines and Builder Projects
- Agent harness benchmarks just moved. Fable 5.1's Terminal-Bench dominance means existing benchmarks need recalibration.
- Cache economics flipped. The 75% cache read price cut changes the optimal architecture for long-running agents.
- Output verbosity is now a cost variable. The 1.7x output increase means agent loops need tighter prompt engineering.
- One model, two personas. Prompt engineering becomes the differentiator when weights are shared.
The Signal
The bottleneck in AI has shifted from "can the model do it?" to "how do we build systems that use it well?" — and then again to "how do we make it economical?" Fable/Mythos 5.1 answers the first question. The prompt engineering and infrastructure challenge is now the hard problem.