On CyberGym it scores 88.4%, ahead of OpenAI's gated GPT-5.5-Cyber at 85.6%. Post-trained from DeepSeek V4.1 Flash for offensive cyber performance.
General coding models are trained to build things. Hinoki is trained to find the ways they break.
Time to start hacking your code before someone else does.
Hinoki is an offensive security model, trained to break software. It is now on AQ, the first platform to offer it, pointed at your own code.
Devin CLI @cognition now runs in AQ.
Cognition's terminal agent, on an always-on machine, in its own git worktree, with your team watching the same live session.
6 months of Claude Max 20x, on us.
We're expanding Claude for Open Source to more of the community.
If you're a maintainer, a core contributor, someone landing PRs across the ecosystem, or someone keeping a critical package alive, apply today!