python-genai v3.2.0

Version 3.2.0 of python-genai, Google's official Python library for its Gemini models, adds one call that matters: `client.decisions.create`, the Python entry point to the Gemini Decisions API that Google announced the same day. The release notes say only that the Decisions API was added to the SDKs; Google's announcement explains what it is for.

The Decisions API answers questions that have a fixed set of possible answers instead of writing free text. A developer sends one input, which can be text, images, audio, video or a PDF, along with up to 1,000 questions about it: yes-or-no questions, multiple-choice questions with up to 255 options, or a grade against an ordered rubric. The underlying model, Gemini Decision 1, scores the options in a single pass rather than generating an answer word by word, and returns each answer with a confidence between 0 and 1, plus a probability for every option on multiple-choice questions, Google says.

That shape suits the small judgments that sit inside larger AI applications, such as deciding which model should handle a request, which tool an agent should call, or whether a message breaks a policy. Because every answer arrives with a probability, the calling code can act on confident answers and send doubtful ones to a slower model or a person. Google puts end-to-end latency at 150 to 250 milliseconds and the price at $0.10 per million input tokens, with no charge for output, and claims the lowest calibration error, meaning its confidence scores most closely match how often it is right, across its own benchmark suite. Those figures come from Google and have not been checked independently, and the announcement says support in Google Cloud is still to come.