Introducing Gemini Decision 1: A Model That Picks Answers Instead of Writing Them

KDnuggets
Read full story· 6 min
Google launched Gemini Decision 1, a model designed to answer closed questions with probabilities instead of generating text. It supports multiple input types and offers cost-efficient API pricing with no charge for output tokens. The model targets decision-making tasks in agent stacks, providing confidence scores for each answer.

What happened

Google released Gemini Decision 1 last week through the new Gemini Decisions API. This model is designed to answer closed questions by providing probabilities for each answer, rather than generating text. It can process various input types, including text, images, video, audio, and PDFs, and is priced at $0.10 per million input tokens with no charge for output tokens. The model is already being used in several Google products to support decision-making tasks.

Key details

  • Gemini Decision 1 can handle up to 262,144 input tokens per request.
  • Users can ask up to 1,000 independent questions per input, including yes/no, multiple choice, and scoring questions.
  • Each answer includes a confidence score between 0.01 and 0.99.
  • API pricing is $0.10 per million input tokens, with no cost for output tokens.
  • A free tier is available for prototyping in Google AI Studio.

Why it matters

Teams building AI-powered systems can now use a dedicated model for decision tasks at a lower cost, as they are only billed for input tokens and not for output. This change reduces expenses for companies that previously used more expensive generative models for similar tasks. Developers and businesses can process large volumes of decision queries more efficiently, and the confidence scores allow them to handle uncertain cases more reliably.

More on this story


More in Machine Learning