OpenHands

<@U06PEF04LEQ> <@U08H0CRU1V4> are we on <https://www.anthropic.com/news/claude-fable-5-mythos-5>? I think we should have access

Also <@U08H0CRU1V4> before running the full eval check to make sure that it's not refusing all your requests <https://x.com/KLieret/status/2064427028661846070|https://x.com/KLieret/status/2064427028661846070>

Hmm, and they are deliberately hobbling the model to make it not work well on developing competing products, which seems not great: <https://x.com/eliebakouch/status/2064399902684139852>

<@U06PEF04LEQ> that is very unfortunate, has anyone tested the boundaries of this new guardrail?

The thing is that you can’t test the boundaries of the new guardrail because they don’t even tell you if they silently nerfed the model’s performance

(although if you tried very hard you could probably find out)

Welp, due to the performance drop from opus 4.6 -&gt; 4.8, I already cancelled my max subscription. I’ve been back on OpenAI :melting_face:

Yea, I'm curious on the OH benchmarks - the fact that it's double the price of Opus and they removed the security aspects which made Mythos interesting makes it a hard sell

Currently doing smoke tests on Fable 5.
Don't have access to Mythos 5.