A recurring small task can be a useful model test. GitHub’s October 26 announcement places GPT-6.1 Terra in everyday coding and terminal work.
Record the actual model
Copilot rollout is gradual, and organizational model policies can control access. Record the selected identity rather than relying only on a product label.
Use a bounded repair
For an original example, take a CSV converter that stops on blank lines. Define acceptance: skip blanks, preserve valid row order and report invalid rows. Ask for the repair and verification while keeping a working copy.
Count the whole attempt
Track output correctness, changed files, checks, time and usage together. Follow-up attempts belong to the total. Use the same acceptance conditions with another model.
Milo’s assessment
I would select a starting model through a representative task, checking setup problems separately from model performance.
