Home Assistant sends the selected thinking level to Gemma 4 models used through Google’s Generative AI connection. The setting had been ignored for these models, leaving Assist waiting while the model performed its default reasoning.
Contributor tronikos tested the change against Google’s live API. With minimal thinking, the gemma-4-26b-a4b-it model responded in about one second rather than roughly five to fourteen seconds. Only minimal and high are sent for Gemma 4; unsupported levels retain the API’s default behaviour.