The model targets coding, computer vision, and scientific research
Anthropic's Fable 5 is out – the first Mythos-class model the public can use. A few things I found notable in the blog post, beyond it seeming like a very impressive model (I'm excited to try it out!):
Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.
Its capabilities exceed those of any model we’ve ever made generally available.
A new data retention policy: 30 days of all traffic on Mythos-class models, wherever they're accessed, including third parties (e.g. AWS Bedrock). Not used for training, human access logged, deleted after 30 days. Meant to thwart misuse that a classifier for a single conversation would miss.
They're setting up trusted-access programs for cyber + bio. Basically, a safeguarded model for everyone (Fable 5) – downgraded to Opus 4.8 if classifiers fire – the full-capability version for vetted users (Mythos 5).
Harder to jailbreak than before: 1,000+ hours of external red-teaming turned up no universal jailbreak. Though the UK AISI "has made progress towards one" in a brief testing window. Maybe AISI will do it again?
They've also clearly put work into vision, where Claude has historically been poor. Fable 5 rebuilt a web app's source code from screenshots alone, and beat Pokémon FireRed with a minimal vision-only harness.
The model targets coding, computer vision, and scientific research
Anthropic's Fable 5 is out – the first Mythos-class model the public can use. A few things I found notable in the blog post, beyond it seeming like a very impressive model (I'm excited to try it out!):
Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.
Its capabilities exceed those of any model we’ve ever made generally available.