Back to posting here. Sorry it's been a while.
Most of us reading this post will never get to use Mythos, but luckily Anthropic decided to release a version with a ton of safety guards (Fable)
The benchmarks look strong, especially the jumps in vision and code.
Curious how Gemini 3.2 and the next GPT models stack up.
Working in the Risk & Compliance space, my daily prompts deal with fraud and money laundering. With the new safety filters, and since I'm not in Anthropic's special program, those exact queries may get rerouted to Opus 4.8 or blocked outright.
The tool built to catch bad actors might end up flagging the people fighting them.
I broke down the full release, press notes, and system card in a new 20-minute video