The model that reasons about your data is the single most important component in our product. It decides how a connection gets built, how a transformation is shaped, how a mapping resolves. So when we say we changed it, that is not a footnote in a release. It is a deliberate choice about the part of the platform that matters most.
This release moved our reasoning layer to Anthropic Claude, and it gave customers the option to bring their own key.
Why Claude
We evaluated the reasoning, not the logo. Our work is not casual chat. It is reading an API specification and understanding how its pieces relate, holding a lot of context about a customer’s data structures at once, and proposing steps that have to be correct because a deterministic engine is going to execute them. On that kind of work, Claude reasoned better and held context better. That is the whole reason.
It also shipped alongside the agentic copilot that builds, validates, and refines pipelines step by step, with a person able to see and modify everything along the way. A better reasoning layer makes that loop noticeably more reliable.
Why bring-your-own-key matters more than it sounds
The release also added bring-your-own-key, so a customer can run reasoning through their own Claude instance. This is not a convenience feature. It is a statement about whose relationship with the model provider this is.
If you bring your own key, the model relationship is yours, on your terms, under your agreements. Combined with the boundary we hold everywhere else, where the model reasons on abstracted context and raw data stays inside the platform, it means an enterprise can adopt AI here without inheriting a dependency it did not choose.
The principle
Pick the best reasoning available, and do not lock customers into how they consume it. That is the posture, and this release is where it became concrete. More on the boundary and controls on the security page, and on the architecture on the platform page.