https://xcancel.com/AISecurityInst/status/204986822774056589...GPT 5.5 appears to have matched Mythos Preview on the UK government AISI "The Last Ones" benchmark. Quoting from the @AISecurityInst thread:
"A key question after our evaluation of Mythos Preview earlier this month was whether its performance was a one-off. GPT-5.5 - a different model, from a different developer - achieving similar results suggests this is part of a broader trend in AI cyber capabilities."
