AI, in your work.

What GitHub announced

The October 19 announcement deprecates six models across most Copilot experiences, including Chat, inline edits, ask and agent modes, and code completions. Manual removal is unnecessary.

These were the suggested replacements in that announcement:

Availability: a listing does not guarantee universal access

The supported models documentation classifies all four alternatives as generally available. However, it still lists Grok 4.5 as available and omits it from the retirement history. This discrepancy prevents describing its retirement as uniform and fully clarified.

The sources consulted also do not enumerate every exception to “most” experiences. The announcement alone therefore cannot establish the status of every model and client combination.

What to review in access settings

According to GitHub’s configuration guide, access depends on the plan, client, and organization or enterprise restrictions. Owners can enable or disable models for Copilot Business and Enterprise users. Free and Student users have access through automatic model selection only.

In VS Code, the Chat model is changed using the dropdown at the bottom of the panel. This choice does not change the model for inline suggestions, as the model switching guide explains.

As an operational recommendation, Nexo suggests recording the experience being used, the model shown, and the applicable policy separately. This short record helps distinguish an access problem from a decision about which alternative to adopt.

Interpretation: replacements need their own acceptance criteria

Nexo’s editorial interpretation is that a provider’s recommendation offers a starting point for evaluating an alternative. It does not demonstrate that the replacement will preserve the behavior a team requires.

A useful review can begin with model references in instructions, training materials, and internal procedures. The team can then define an acceptable response: respecting an interface, preserving existing tests, or explaining a change precisely enough.

If the previous model is no longer accessible, the comparison can use explicit requirements and previously retained outputs. It should be described as an acceptance evaluation, without presenting it as a simultaneous test of the models.

A synthetic evaluation example

The following exercise is a Nexo proposal; it has not been run and does not represent any model’s results. It uses a fictional repository so that the example does not depend on private code.

• Task: fix a function that adds amounts and fails when given an empty list.
• Criteria: return zero for an empty list, preserve the public signature, and add tests for normal and empty inputs.
• Record: note the model, client, instructions, supplied context, tests passed, and human interventions required.