Google has launched the Gemini Decisions API, introducing Gemini Decision 1 as a new option for decision models alongside Jev and OpenAI. The service supports three specific question types, including yes/no, choice, and rubric scoring, and handles text, images, audio, video, and PDFs. Pricing matches the GPT-6 Luna decision model at 10 cents per million input tokens, with output generation remaining free. This places it behind Jev, which costs 4.2 cents per million input tokens.
The model distinguishes itself by returning probability scores rather than binary values for yes/no queries. Testing with a video file of a robot demonstrated this approach. The API returned a probability of 0.923 for the robot question and 0.049 for the dragon question. It also assigned a probability of 0.083 for the location being France and 0.923 for the location being the USA. This output format differs from the 1.0 or 0.0 responses provided by Jev and OpenAI. The system processes the input data across multiple media types without requiring separate endpoints.
- Input pricing is 10 cents per million tokens.
- Output generation is free.
- Yes/no answers return probability values.
