Verified, then logged.
Sun 22 Nov 2026 · 00:21 KST / 15:21 UTC한국어 | ENSearch /

01 / AI · Claim vs. Check · Analysis

Google launches API to score predefined choices

Shared input is billed once per request, while Google's latency figures apply to specified input lengths and question counts.

Over an out-of-focus clinician’s shoulder, monitors in a quiet hospital consultation room display generic patient records beside a clinical summary and an insurance authorization draft. AI-generated image

Google announced the Gemini Decisions API at $0.10 per million input tokens and $0 for output tokens for developers.

According to Google's announcement, the API scores predefined choices using Gemini Decision 1 and returns probabilities and confidence scores. Its purpose is to assess options rather than generate an answer.

Google described a structure that lets multiple questions share the same input. The distinction between how that input is billed and the conditions behind the published latency figures is central to the announcement.

Shared input billing

Google said shared input is billed once per request, regardless of how many questions use it. Instructions and assessment criteria added for each question count towards that question's tokens.

The announcement specifies no separate charges for audio transcription, which turns speech into text, or optical character recognition, which reads text from images.

Google's company-published benchmark figures put the Decisions API at nearly five times cheaper than Gemini 3.5 Flash-Lite through the Interactions API across their shared evaluation suite. That comparison describes costs on the suite.

Input limits and response latency

Google specified limits of 256K tokens of shared input and 1,000 independent questions per request. Supported inputs include text, images, audio, video and PDFs.

Google's company-published performance figures give end-to-end latency of 150–250 ms for inputs of up to 16K tokens. Latency remains nearly flat across 1 to 100 questions per request within that range.

The announcement describes using probabilities and confidence scores to send ambiguous decisions to a human reviewer. The results centre on which option is selected and how confident the assessment is.

Routing queries to different models

Google said Gemini Notebook uses the API to route questions by model tier and reasoning level. This sends everyday queries and complex questions to different models.

The announcement positions the API as a decision tool that selects a route before another model generates an answer. Google also announced forthcoming support in Gemini in Google Cloud, but did not disclose a start date.