python-genai v3.1.0

python-genai is Google's official Python library for its Gemini models, and version 3.1.0 is a small feature release. Its headline addition, per the release notes, is the model name `gemini-omni-1.2-flash` for Gemini Omni 1.2 Flash, the video-generation model Google announced separately. The release makes it the `gemini-api video` default and adds it, with `gemini-4-carbon`, to the model list of the Interactions API, Google's interface for calling models and agents in which each call is an interaction that can be stored and continued later.

The change most likely to affect a bill is a new `proactive_video` setting for the Live API, which runs real-time sessions in which an app streams audio and video to the model. It requires the existing `proactive_audio` setting, which lets the model stay silent when speech is not meant for it, to be on as well. According to the field's description in the published 3.1.0 package, enabling `proactive_video` means the system might start a response on each image it receives, so the number of responses follows the rate of incoming frames, and the description warns this can significantly increase cost. An app that streams camera frames should treat turning it on as a cost decision.

One smaller addition applies to the Gemini Enterprise API: embedding responses now carry usage metadata, including the traffic type, and a `truncated` flag that marks input text cut for exceeding the maximum length. The release lists no breaking changes, but projects still on 2.x must first absorb version 3.0.0, which moved the default Gemini Developer API version from `v1beta` to `v1` and replaced the default HTTP client `httpx` with `httpx2`.