# Google’s Gemini Decisions API Compared With Microsoft Decision 1, Jev, and OpenAI

> Source: <https://pub.towardsai.net/googles-gemini-decisions-api-compared-with-microsoft-decision-1-jev-and-openai-85bcebb8125d?source=rss----98111c9905da---4>
> Published: 2026-11-18 02:53:26+00:00

**Key takeaway:** Gemini Decisions API accepts text, images, audio, video and PDFs and answers up to 1,000 independent questions per request at $0.10 per million input tokens with no output-token charge, versus 400 questions per request for OpenAI's Decisions API and text/JSON-only input for Microsoft Decision 1.

*Illustration generated with AI.*

On November 11, Google released the **Gemini Decisions API** and the model behind it, **Gemini Decision 1**. You give it evidence and a fixed question. It can estimate the probability that a condition is true, choose from options you supply, or score something against a rubric. An app can use those answers to route a request or flag it for review. [Google’s launch announcement](https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/) · [Decisions API documentation](https://aistudio.google.com/docs/decisions-api)

The part that caught my attention was the input. Gemini can evaluate text, images, audio, video, and PDFs. You can ask up to 1,000 independent questions about the same input in one request, within the context limit. Google AI Studio offers a playground and a free tier for prototyping. The paid rate is $0.10 per million input tokens, with no output-token charge. [Gemini Decision 1 model card](https://aistudio.google.com/docs/models/gemini-decision-1) · [Gemini API pricing](https://aistudio.google.com/docs/pricing#gemini-decision)

**Microsoft Decision 1** accepts text or JSON through Microsoft Foundry, with a listed rate of $0.042 per million input tokens. When I tried it for UI checks, I struggled with the cases where the evidence I needed was visual. For decisions based on structured state or text, Foundry still appeals to me because it also offers models from OpenAI and Anthropic. [Microsoft Decision 1 docs](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision) · [Microsoft pricing announcement](https://commandline.microsoft.com/microsoft-decision-1-model-foundry/) · [Foundry model catalog](https://ai.azure.com/catalog/models/)

**TypeSafe’s Jev** also lists $0.042 per million input tokens. I couldn’t attach images in the setup I tried, and bringing in a separate provider for one model felt like a lot of work for a company. TypeSafe now offers experimental image input, and Jev 1.14 can return a bounding box around a target in an image. OpenRouter also offers Jev and Gemini Decision 1, which gives teams already using it another way to compare them. [TypeSafe’s release](https://typesafe.ai/blog/jev-1-14) · [Jev pricing](https://help.typesafe.ai/articles/5187202169-how-much-does-jev-cost) · [Jev on OpenRouter](https://openrouter.ai/typesafe/jev-1.14) · [Gemini Decision 1 on OpenRouter](https://openrouter.ai/google/gemini-decision-1)

**OpenAI’s Decisions API** is generally available with GPT-6 Luna. It accepts text, images, and JSON and supports up to 400 questions per request. An image can be shared across the request or attached to an individual question. The API also supports prompt caching, with a listed rate of $0.10 per million uncached input tokens before applicable adjustments. All four APIs list zero charges for output tokens. [OpenAI Decisions API documentation](https://developers.openai.com/api/docs/guides/decisions)

*Author-created chart. Direct media inputs listed in the provider documentation on November 17, 2026.* [Google](https://aistudio.google.com/docs/decisions-api#input) · [Microsoft](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision) · [TypeSafe](https://docs.typesafe.ai/images) · [OpenAI](https://developers.openai.com/api/docs/guides/decisions)

The **Artificial Analysis Decision Index v2.1** combines six evaluations covering document decisions, tool selection, knowledge, reasoning, and consistency when questions are rephrased. It weighs how confidently models answer alongside whether they get the answer right. Its tasks use text, so these scores tell us about decisions on that test suite. Image, audio, and video capability still needs its own evaluation. [Decision Index](https://artificialanalysis.ai/evaluations/artificial-analysis-decision-index) · [Methodology](https://artificialanalysis.ai/methodology/decision-index)

**Index score, time per task, and estimated cost per million tasks**

*Artificial Analysis snapshot, November 17, 2026. Scores are rounded. Time combines each evaluation’s median using the index weights. Cost is estimated for the benchmark’s tasks at listed prices. The Microsoft timing is for OpenRouter.* [Gemini](https://artificialanalysis.ai/decision-models/gemini-decision-1) · [Microsoft](https://artificialanalysis.ai/decision-models/microsoft-decision-1) · [Jev](https://artificialanalysis.ai/decision-models/jev-1-14) · [OpenAI](https://artificialanalysis.ai/decision-models/gpt-6-luna-decisions)

*Author-created chart. Time per task and estimated cost per million tasks for the routes measured by Artificial Analysis. Lower is better on both measures.* [Decision Index v2.1](https://artificialanalysis.ai/evaluations/artificial-analysis-decision-index)

Jev and Gemini rounded to the same score. Jev took less than half the time per task and cost less than half as much on this suite. For my own work, Gemini is still the one I’d try next when the evidence is an image or a recording. I can experiment in AI Studio alongside the other Gemini models, which makes it easier for me to justify trying a decision model in the first place.

[Google’s Gemini Decisions API Compared With Microsoft Decision 1, Jev, and OpenAI](https://pub.towardsai.net/googles-gemini-decisions-api-compared-with-microsoft-decision-1-jev-and-openai-85bcebb8125d) was originally published in [Towards AI](https://pub.towardsai.net) on Medium, where people are continuing the conversation by highlighting and responding to this story.
