I've seen various claims to this (eg [1][2][3]) but nobody reall knows. These may all come from one uunsubstantiated claim. It is I think widely accepted that M

I've seen various claims to this (eg [1][2][3]) but nobody reall knows. These may all come from one uunsubstantiated claim. It is I think widely accepted that Mythos is ~10T parameters. I've seen figures that suggest GPT-4 was 1.8T parameters and cost upwards of $100 million to train (also unsubstantiated), in which case the Mythos figure might be inflated and also include development costs. So who really knows? [1]: https://www.softwarereviews.com/research/claude-mythos-preview-and-project-glasswing-what-it-and-security-leaders-need-to-know-now https://www.softwarereviews.com/research/claude-mythos-previ... [2]: https://x.com/duttasomrattwt/status/2041903600516133016 https://x.com/duttasomrattwt/status/2041903600516133016 [3]: https://www.forrester.com/blogs/project-glasswing-the-10-consequences-nobodys-writing-about-yet/ https://www.forrester.com/blogs/project-glasswing-the-10-con...