Cursor (Anysphere)Cursor completion
Cursor Tab
The Tab model learns from accepted and rejected edit suggestions. Cursor’s training report describes frequent checkpoint updates with online reinforcement learning.
What the release provides
Tab can propose changes across multiple lines and files. Its current help page says predictions use recent edits, nearby code and linter errors. It is a separate workflow from selecting a Composer agent model.Current Cursor Tab workflow
How to use it well
Practical guidance. Read the proposed edit before accepting it. For a repeated rename or refactor, inspect all affected references rather than accepting a sequence only because each local suggestion looks plausible.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Provider product |
| Access | Available through Cursor’s Tab completion interface; no standalone public model ID is documented here.Current Cursor Tab workflow |