Google Launches Gemini Decisions API With Up To 1,000 Questions Per Request
Google’s Gemini Decisions API handles up to 1,000 questions per request at the same base input price as OpenAI’s Decisions API.
Google introduced the Gemini Decisions API in a company blog post on 11th November, giving developers a way to ask software to choose from a set of predefined answers. The service uses the Gemini Decision 1 model. Developers can access it in Google AI Studio using the Gemini API. OpenAI made its Decisions API generally available on 22nd October after a public beta.
Google says its service can evaluate up to 1,000 independent questions in one request. Its documentation sets a combined limit of 262,144 tokens across the shared material and all questions. Tokens are the units used to measure and bill model input. Supported inputs include text, images, audio, video and PDFs. OpenAI’s October announcement says its GPT-6 Luna service can handle up to 400 questions and accepts text, images and structured data in JSON.
Both services let developers set the answers in advance, such as choices for where to send a support request. Google’s model can also estimate the probability of a yes or no condition or grade an input against an ordered scale. It supplies confidence scores that applications can use to send uncertain cases for human review. Bored Node’s report on agent wallet approvals examined how separate approval controls can matter when a software decision leads to a financial action.
The Same Base Input Rate
The companies’ announcements both list a base price of $0.10 per million input tokens with no charge for output. Google’s pricing documentation says it bills shared material once per request, plus the tokens for each question. OpenAI separately lists $0.01 per million cached input tokens for previously processed material. Its current pricing guide also lists charges for writing to the cache, premiums for regional processing and multipliers for long inputs. Those conditions can change the bill for a particular request.
Google reported median end-to-end times of 150 to 250 milliseconds in its tests on inputs of up to 16,000 tokens. It said timing was nearly flat between one and 100 questions. OpenAI’s October post reports speeds of up to 10 times those of its own GPT-6 Luna through the Responses API. The companies reported different speed measures. Our coverage of GPT-6.1 Terra’s benchmarks examines how a model’s reported speed, cost and reliability depend on the task and test conditions.
