Google launches Gemini Decisions API for scoring up to 1,000 questions per request
For a model-risk officer at a regional bank
Google introduced the Gemini Decisions API, powered by its new Gemini Decision 1 model, on Nov. 11, according to a post on the Google blog by Google DeepMind's Min Choi and Felix Yu.1
Google said the endpoint is built for bounded choices such as tool selection, policy checks, model routing and context grading, and that Gemini Decision 1 scores options in a single forward pass instead of generating answer text token by token.1 Google's documentation says generating and parsing text adds latency and brittle parsing logic when code needs a judgment to branch on.2
One shared multimodal input across text, images, audio, video and PDFs (up to 256K tokens) can carry up to 1,000 independent questions, and Google gave a call recording paired with account metadata and a scanned PDF claim with damage photos as examples.1
Questions can be yes/no, choice (2 to 255 options) or ordered rubric score, and each answer carries a confidence score from 0 to 1 that code can use to route ambiguous cases to a fallback or human reviewer, according to Google.12
Google priced Gemini Decision 1 at $0.10 per 1M input tokens and $0 for output tokens, billed once for the shared input per request, with end-to-end latency of 150-250 ms and a no-cost API tier in Google AI Studio for prototyping.1
Google said in its blog post that on benchmarks spanning agent routing, reasoning, ranking, and multimodal understanding, Gemini Decision 1 leads across the suite, with the lowest calibration error.1
Google's own examples are low-stakes uses such as query routing and emoji selection, and the sources give no evidence of accuracy on regulated decisions like claims or complaints. A bank testing it for those would need to validate confidence scores on its own data before relying on a threshold, and cheap per-question pricing makes it easy to add checks faster than they get validated.
Google said ecosystem integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM and others, with support coming soon to Gemini in Google Cloud, and did not give a date.1
Our take. Google sells confidence scores at $0.10 per 1M tokens and grades their calibration on its own benchmarks. Your scorecards never got to grade themselves. This one should not either.
Action. Model risk: Add LLM decision endpoints like gemini-decision-1 to the model inventory and require calibration tests on bank data before production use.
What our records say
Gemini Enterprise is Google Cloud's AI assistant and agent platform, launched in October 2025, and it sells banks a per-seat AI app for employees plus an agent-building platform, with a financial-services edition since August 2026.1 Strategy: Gemini Enterprise for Financial Services is in preview (Aug 25, 2026), a financial-services edition for capital markets and corporate banking with Deutsche Bank as a design partner for its Financial Research agent.2
From our records
- Gemini Enterprise (Google Cloud)vendor · banks using it (11) Wells Fargo, BNY, Citi · alternatives Microsoft Copilot, ChatGPT Enterprise, Claude
Sources
- Primary sourceGemini Decisions API: Fast multimodal decisions Google
- Primary sourceDecisions API - Gemini API | Google AI Studio Google
- Also reported byGoogle joins the decision-model wave with Gemini Decision 1, scoring 1,000 questions in a single pass The-decoder