Google launches Gemini Decisions API for fast, calibrated multimodal choices

Google AI announced on 11 November 2026 the Gemini Decisions API, a new endpoint that delivers rapid, probabilistic decisions across text, images, audio, video and PDFs, powered by the Gemini Decision 1 model [1].

Purpose‑built decision primitives

The Gemini Decisions API is designed for bounded‑choice tasks such as tool selection, policy checks, model routing and content grading. Users submit a single multimodal input—up to 256 KB of tokens—and can attach up to 1,000 independent questions in one request. The API returns probabilities and confidence scores for each query without generating answer text token‑by‑token, which reduces latency and eliminates output‑token billing [1].

Performance and cost

According to Google, Gemini Decision 1 scores options in a single forward pass, delivering end‑to‑end latency of 150–250 ms and a median compute latency of 50–100 ms, even when handling the maximum 1,000 questions per request [1]. The service is priced at $0.10 per 1 million input tokens with zero charges for output tokens, making it nearly five times cheaper than the Gemini 3.5 Flash‑Lite model accessed via the Interactions API [1]. Benchmarks covering agent routing, reasoning, ranking and multimodal understanding show Gemini Decision 1 achieving the lowest calibration error across the suite [1].

Multimodal flexibility

Built on Gemini’s native multimodal architecture, the model can process text, images, audio, video and PDF documents in a single forward pass. Users may combine modalities—for example, pairing a call recording with account metadata or a scanned claim PDF with damage photos—and can adjust image and video resolution to balance visual detail against token cost [1]. The API supports three decision types: - noul – yes/no probability with optional true/false criteria. - choice – selection from 2 to 255 options, returning the winner and normalized probabilities. - score – ordered rubric grading, delivering a continuous weighted score and per‑level probabilities [1].

Early internal adoption

Google teams are already leveraging the Decisions API as a low‑latency control plane. Gemini Notebook uses it to route queries to the optimal model tier, achieving up to 40 % lower latency while reserving deeper models for complex tasks [1]. The rapid‑canvas experiment Stitch generates playable wireframes in seconds and applies voice or tap edits instantly, before handing off to the full design tool [1]. Google Labs’ CC project selects emoji reactions from a palette of more than 200 options in fractions of a second [1]. Additionally, Google Search incorporates the API as a safety layer to detect adversarial content within the latency envelope [1].

Ecosystem rollout

Integrations are being added across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode, with the google‑genai SDKs already supporting the endpoint. Support for Gemini in Google Cloud is slated for a near‑future release [1].

Availability

Developers can obtain an API key through Google AI Studio and start prototyping with a no‑cost tier. Demo applications such as Doorstep Parcels, Speed Sketch and Find the Zoo Keeper illustrate typical use cases [1].

Google’s Gemini family of models underpins a range of AI services, with Gemini Decision 1 representing a specialized variant focused on closed‑set decision making rather than open‑ended generation. The Decisions API extends this capability to developers, offering a single‑pass, multimodal evaluation that contrasts with traditional token‑by‑token generation approaches used in many large language models [1].

For Indian enterprises and developers, the Gemini Decisions API provides a way to embed fast, calibrated decision logic into applications such as fraud detection, content moderation, customer support routing and automated grading, all while keeping compute costs low. The sub‑second response time aligns with real‑time user experiences expected in fintech, e‑commerce and digital services across India.

If adopted broadly, the API could reduce the need for custom classifier training, lower infrastructure spend for high‑volume decision workloads, and accelerate product development cycles in sectors that rely on rapid multimodal assessment, such as insurance claim processing, media moderation and financial risk scoring.

The service remains tied to Google’s cloud ecosystem, raising concerns about vendor lock‑in and data residency for Indian firms with strict compliance requirements. Calibration performance, while benchmarked as strong, may vary on domain‑specific data, and the pricing model, though cheaper than alternatives, still incurs costs that need careful budgeting.