Gemini Determination 1 scores choices with out token-by-token decoding, preserving median latency 50-100 ms of mannequin compute (150-250 ms end-to-end), throughout multimodal inputs and contexts as much as 16K tokens, remaining almost flat from 1 to 100 questions per request. With zero output tokens billed, Gemini Determination 1 with the Choices API is sort of 5x cheaper throughout the shared benchmark suite than Gemini 3.5 Flash-Lite with the Interactions API.
Native multimodal analysis throughout three determination varieties in a single move
Constructed on Gemini’s native multimodal structure, Gemini Determination 1 can course of textual content, photographs, audio, video, and PDFs straight in a single ahead move. You may mix modalities in a single shared enter, comparable to pairing a name recording with account metadata or a scanned PDF declare with injury photographs, and tune decision on picture and video inputs to manage visible element and token value.
You may consider a number of questions towards that shared enter in a single request, mixing sure/no checks, a number of alternative, and rubric scores. Each reply features a confidence rating (from 0 to 1) so your code can route ambiguous instances to a fallback or human reviewer:
- noul (Sure/No): Estimates the chance {that a} sure/no situation holds, with optionally available true and false description standards.
- alternative (A number of Alternative): Selects from 2 to 255 selections, returning the winner and normalized possibilities for each candidate.
- rating (Ordered Rubric): Grades inputs throughout an ordered rubric (beginning at 0), returning a steady weighted rating and per-level possibilities.
The enter subject makes use of the identical Content material block format because the Interactions API. Right here is find out how to test a photograph of a breakroom counter, along with a textual content word, towards all three query varieties in a single request:
