Instruction Following
How precisely a model executes a user's explicit and implicit constraints — a core axis of coding precision and reliability
What is instruction following?
Instruction following is the model's ability to execute a user's request faithfully — without quietly altering or skipping parts of it. Adherence to fine-grained constraints such as "keep the signature, no side effects" is what turns this property into reliability.
Why is it decisive in coding?
In coding, "what must not change" is as important as "what must change." Models with weak instruction following modify code beyond the requested scope and produce regression bugs. Strong models ask back or preserve when an instruction is ambiguous, instead of inferring constraints that were never specified.
How is it compared?
There is no single canonical benchmark. Practitioners triangulate from release-note callouts, secondary evaluations such as IFEval, and observed regression rates in real usage. Differences become more visible under non-English PRDs and ambiguous natural-language requests.