Google’s Gemini Decisions API interface lets developers submit text, images, audio, video, or PDFs for instant decision scoring.
GOOGLE LAUNCHES GEMINI DECISIONS API FOR FAST MULTIMODAL CHOICES
*Google unveils a new endpoint that scores up to 1,000 multimodal questions in a single pass. The service promises sub‑second latency, calibrated confidence scores, and a cost model that undercuts existing Gemini offerings.*
Google announced the Gemini Decisions API on November 11, 2026, positioning it as a fast, calibrated decision engine for developers. The new endpoint handles up to 1,000 yes/no, multiple‑choice, and rubric‑scoring questions in a single pass across text, images, audio, video, and PDFs. By scoring options instead of generating text, the service delivers sub‑second responses and a cost structure that undercuts existing Gemini models. Product Manager Min Choi and Principal Scientist Felix Yu highlighted the API’s role as a low‑latency control plane for routing, grading, and policy checks, promising a new layer of efficiency for AI‑driven applications.
What does the Gemini Decisions API actually do?
The Gemini Decisions API, powered by the Gemini Decision 1 model, accepts a single multimodal input—text, images, audio, video, or PDF—up to 256K tokens. Developers can attach up to 1,000 independent yes/no, multiple‑choice, or rubric‑scoring questions. Instead of generating answer text token by token, the model scores every option in one forward pass and returns probabilities and confidence scores between 0 and 1. The API is accessed through Google AI Studio with standard Python, JavaScript, or REST guides. It is positioned as a low‑latency control plane for routing, grading, and policy checks.
Performance claims: speed, cost, and calibration
Google reports end‑to‑end latency of 150‑250 ms per request, with median compute latency of 50‑100 ms, staying flat from one to 100 questions. Pricing is $0.10 per 1 million input tokens, with zero output tokens billed, making it nearly five times cheaper than Gemini 3.5 Flash‑Lite on the shared benchmark suite. Benchmarks across agent routing, reasoning, ranking, and multimodal understanding show Gemini Decision 1 leading the suite with the lowest calibration error, indicating more reliable confidence scores than competing models.
Multimodal capabilities and decision types
Built on Gemini’s native multimodal architecture, the API processes all five data types in a single forward pass. Users can blend modalities—such as a call recording with account metadata or a scanned claim PDF with damage photos—and adjust image or video resolution to manage token cost. The service supports three primitive decision types: "noul" for yes/no probability, "choice" for 2‑255 options with normalized probabilities, and "score" for ordered rubric grading starting at zero. Each response includes a confidence score, enabling automated fallback or human review for ambiguous cases.
How are Google teams using the Decisions API?
Product teams inside Google have adopted the Decisions API as a real‑time router. Gemini Notebook, for example, uses the API to instantly classify intent and direct queries to the optimal model tier, reducing latency and eliminating the need for custom classifiers. The low‑latency control plane lets engineers ship features by defining option sets rather than training bespoke models or waiting on generative calls. Early internal use demonstrates the API’s promise for dynamic model routing and policy enforcement across Google’s AI products.
Frequently asked questions
How many questions can the Gemini Decisions API handle per request?
The API can process up to 1,000 independent yes/no, choice, or rubric‑scoring questions in a single request.
What is the latency and cost of using the Gemini Decisions API?
It delivers 150‑250 ms end‑to‑end latency and costs $0.10 per 1 million input tokens, with zero output tokens billed.
Which decision types does the Gemini Decisions API support?
It supports three types: "noul" for yes/no probability, "choice" for 2‑255 options with normalized probabilities, and "score" for ordered rubric grading.
If the early internal adoption signals anything, the Gemini Decisions API could become the backbone of real‑time AI decision making across the industry. Its blend of speed, multimodal flexibility, and transparent confidence scores addresses a long‑standing bottleneck in AI pipelines. As developers integrate the endpoint into routing, compliance, and content grading, the pressure will mount on competing providers to match its latency and pricing. The next wave of AI products may be defined not by generative prowess but by how swiftly they can turn data into actionable decisions.
Sources: Hacker News (https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/)