37% → 69%
on failures a policy recognized, same model.
See the research →
×
docs
research
pricing
blog
token times
events
login →
the token times
/
daily
/
6 nov 2026
OpenAI's cyber model gets a tighter leash
← back to the edition
