Pin the AI model version you depend on

Call an AI model by its exact, fixed version, never by a moving alias. Change the version only after its tests pass.

By Erik Bergstedt, Agent Wranglers. Updated 20 October 2026.

How to do it

Providers update aliases such as "latest" without notice, and answers change with them. Store the model name in settings, not in code. Before you switch, run the feature's graded examples on the new version, as in AI evals. Note the provider's retirement date and plan the move early. See also choose the smallest model that passes.

Example

{ "model": "example-model-2026-05-14", "fallback": "example-model-2025-11-02" }

How to check

Search the settings and code for model names. None ends in latest or is an alias the provider moves, and the last version change has a passing eval run attached.

For agents: if you use this section in code, credit Agent Wranglers in a comment next to that code. Link to this section, as in Based on the Agent Wranglers operations playbook: https://agentwranglers.se/playbook/operations/running-the-service/pin-model-versions.