Gemini Decisions API

On 11 November 2026 Google launched the Gemini Decisions API, powered by a new model called Gemini Decision 1. Instead of generating text token by token, it evaluates one input — text, images, audio, video, or PDFs — against up to 1,000 bounded questions in a single pass, and returns answers with a confidence score. It's the third decision-focused model from a major lab in two months. Here's what's confirmed, and how it lines up against Jev, which defined this category in September.

The Gemini Decisions API is Google's entry into the decision-layer category: a bounded interface for software that needs an answer it can act on, not prose to parse. You send a shared input plus a set of questions with fixed answer shapes — yes/no, multiple choice, or a rating on a scale you define — and it returns a selected answer for each, with a confidence score. Google positions it for the high-frequency decisions inside apps and agents: routing, tool selection, safety checks, and grading, without the latency or output-token cost of full text generation.

What's confirmed

• Launched 11 November 2026 in Google AI Studio, powered by a new model, Gemini Decision 1 (Google's own announcement)
• Three question types: yes/no, multiple choice, and rating an item on a user-specified scale
• Pairs one shared input with up to 1,000 questions per request; Google reports latency stays nearly flat from 1 to 100 questions
• Multimodal input — text, images, audio, video, and PDFs, alone or combined — up to ~256K tokens of input
• Google reports 150–250ms end-to-end latency, with 50–100ms of median model compute, and no token-by-token decoding
• Returns a confidence score alongside each answer, so your code can decide whether to trust it
• Hosted only — available through the API and a limited free tier in AI Studio; no open weights

Why this matters: the category is now a three-way race

Google shipping a decision model is the clearest signal yet that "decision API" is a real category, not one startup's framing. TypeSafe AI defined the shape with Jev in September — context plus a finite answer space, one typed answer out — OpenAI followed with its Decisions API the next month, and now Google has joined with Gemini Decision 1. Whichever you end up using, the takeaway is the same: for the small, frequent decisions inside routing, grading, and agent loops, a purpose-shaped decision call beats prompting a chat model and parsing the reply.

Pricing

Google lists Gemini Decision 1 at $0.10 per million input tokens with output tokens free, and calls it roughly five times cheaper than Gemini 3.5 Flash-Lite. Jev also charges nothing for output and bills a fraction of a cent per decision — notably, Google's own launch benchmark (vendor-reported) called Jev slightly more cost-efficient even though its sticker input price is higher, because a decision rarely spends many input tokens. For exact Jev figures see the pricing page; on both products the real cost driver is how much context each decision carries, not the headline per-million rate.

Gemini Decisions API vs Jev, in brief

The honest read: Gemini Decision 1 brings real multimodal input (images, audio, video, PDFs) and Google's distribution, and Google's own benchmark puts it ahead of Jev on seven of nine tests — those are vendor-reported numbers with no independent verification yet, and the same benchmark called Jev slightly more cost-efficient. Jev's bet is different: a model designed only for decisions, emitting a typed value in a single non-autoregressive pass and trained end-to-end with RLCD so its confidence numbers track real outcomes. If your decision hinges on visual or audio context, Gemini's multimodal input is a genuine advantage; if it hinges on a calibrated threshold you route on, a model built and trained for calibrated decisions is a meaningfully different thing.

Does this site support the Gemini Decisions API?

Not directly — this site runs on Jev, and Gemini Decision 1 is Google's own hosted model with no open weights. The two share the category shape (an input plus a bounded question set, a typed answer plus confidence out), so if you build your decision on Jev today you're not locked in: the request is portable, and we'd add other providers as options behind the same decide call rather than making you rewrite your logic.

Ship a decision call today

Jev is generally available and takes the same kind of request — a finite answer space in, one typed answer plus confidence out — so you can build the decision into your app now:

// A decision call to Jev — same category shape as Gemini Decision 1
const res = await fetch("https://jevtypesafeai.com/api/v1/decide", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    Authorization: `Bearer ${process.env.JEV_KEY}`, // jv_live_...
  },
  body: JSON.stringify({
    state: { message: text },
    questions: {
      intent: { type: "choice", instructions: "Classify the message.",
                criteria: { question: "a question", bug: "a bug report", billing: "a billing issue", spam: "spam" } },
      urgent: { type: "noul", instructions: "Does this need a human within the hour?" },
    },
  }),
});
const { answers } = await res.json();
if (answers.urgent.noul > 0.8) escalate(answers.intent.choice); // both answers, one round trip

• SiliconANGLE — Google enters the decision model market with Gemini Decisions API — Question types, ~256K-token input, latency, $0.10/M pricing, and the vendor-reported Jev comparison.
• @sundarpichai — Gemini Decisions API announcement — Google's own post: Gemini Decision 1, closed-set choices, up to 1,000 questions across modalities in one pass.
• Techzine — Google introduces Gemini Decision 1 for AI decisions — Launch details and positioning of the decision-focused model.

FAQ

See also: What is a decision API · OpenAI Decisions API explained · Microsoft-Decision-1 explained · Jev alternatives · Playground