This has no mention of what happens to the prompt cache, including the "learn more" link.
Knowing Anthropic, it wouldn't surprise me if it will result in a full cache miss/rewrite at fallback, with potentially up to 1M tokens in the context window.
You're paying in full for the "guardrails" embedded in the system prompt, prompt injections, refusals, fallbacks and everything else that may be caused by the service provider.
I must admit I was drawn to Claude because of their (obviously successful with me) publicity stunt about "being too dangerous to make widely available", when Mythos came out.
It's a shame because I was really looking forward to use it specifically to find potential security holes in my own software.
That being said, Opus 4.7/4.8 have been quite useful already, especially for finding things in the harder to test, non-happy paths.
If I recall, Fable 5 is supposed to be basically Mythos which falls back to Opus 4.8 when dealing with cybersecurity. I wonder if that also includes "finding bugs that could lead to security exploits".
Yeah, I was very excited for Fable to come back to use it for work after using Opus 4.8, but now I guess I'm just excited for Sol/Terra/Luna (unless they have the same restrictions)
It's very likely that OAI models will have even more restrictions. Firstly because now they know what feds will do if you don't tune the safety classifiers towards more false positives and secondly, OAI models were always more restrictive than ANT.
Senior Unix/Linux infrastructure consultant with 20+ years of experience,
including SRE roles at Google, Two Sigma, and Reddit.
Available for short engagements (a few days to weeks, flexible scope) covering
server deployment and hardening (Linux, *BSD), DNS, networking,
configuration management (Puppet, Ansible), system health monitoring
(Prometheus, Grafana, Nagios), and backend service development in C,
Go, Perl, or Python — particularly for automation and infrastructure tooling.
Vendor-neutral. Code delivered is yours under a commercial-friendly open-source
license. Available in English or French.
Short-term engagements, no long-term commitment required — $175 CAD/hour.
"You won't be charged Fable prices for rerouted requests. Learn more about how the fallback experience works."
https://www.anthropic.com/claude/fable Under "Safeguards"