What changes with the Gemini Decisions API
On November 11, Google DeepMind presented the Gemini Decisions API, a service that allows hundreds of micro-decisions to be submitted to artificial intelligence within a single request, according to a report by AI4Business. The underlying model, called Gemini Decision 1, can evaluate up to a thousand questions about the same set of information, which may include text, images, audio, video and PDFs.
The difference compared with a traditional generative model lies in the form of the output: the API returns probabilities associated with the requested assessments, rather than explanations in natural language. In practical terms, the user formulates the questions and obtains a confidence measure for each, without a reasoned text justifying the answer. The announced price is $0.10 per million input tokens, with no charges for output tokens.
Why businesses may care
The approach of multiple assessments on a single multimodal input is designed for repetitive verification activities, where the same document or content must be checked against many criteria. Concentrating the questions in one call reduces the number of requests needed and, according to the stated pricing model, eliminates costs related to text generation.
The comparison with Jev, Microsoft and OpenAI is cited by the source as a benchmark, but the available material does not specify the features, prices or performance of the competing services. Likewise, the source notes that some tests measure the model's reliability without detailing the results. It therefore remains to be verified in which business scenarios the returned probability is sufficiently accurate for the required checks.
What remains to be established
The news confirms the launch and the declared technical characteristics, but does not provide information on measured accuracy, actual availability for businesses, or the API's operational limits. Even the calculation of cost-effectiveness compared with traditional generative models depends on unquantified factors, such as the number of assessments actually needed and the quality of the responses.
In the absence of these data, the decision to adopt Gemini Decision 1 should be considered an assessment to be conducted case by case, verifying the results of reliability tests and comparing the cost per million tokens with the expected volume of requests.
