What happened
Google introduced the Gemini Decisions API backed by Gemini Decision 1, which handles closed-set decisions such as yes/no, multiple-choice (up to 255 options) and rubric scoring in a single pass, accepting text, image, audio, video and PDF inputs with 256K context and a confidence score on every answer.
Why it matters
This gives developers a way to get fast, scored answers from a single model call instead of generating text token by token, which Google says enables responses in 150–250 ms end to end.
WHO IT HITSDevelopers building classification, routing, or scoring features into applications can now call a single API that returns a confidence-scored answer in one pass, potentially simplifying pipelines that previously required text generation or multiple model calls.
Summaries like this, every morning.
Gemini Decision 1 originated as an internal Google DeepMind prototype, and a small team shipped the model and API about six weeks later. That short path from prototype to production suggests Google is moving quickly to stake out the decision-model category.
The launch positions Google ahead of Anthropic in releasing a production decision model. At the same time, independent evaluation from Artificial Analysis paints a more mixed picture than Google's own tests, with Decision 1 scoring 66 on the Decision Index—tying Elephant 9B and Jev 1.14 but sitting 7 points behind Drex Max v1.1 and GLiDE v1.2. Its cost and latency also differ from rivals: $345 per 1M tasks versus $103 for Drex Max and $156 for Jev, and an average time per task of 211 ms compared with Jev's 92 ms. These trade-offs may shape how developers choose among decision models for different workloads.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Notion Labs said on November 9 that unauthorized access to contractor Fireworks.ai may have exposed chat recor…
A developer running FLYSTAR pointed the code's base_url at one OpenAI-compatible endpoint and could call GPT-5…
A Kiva engineer published a hands-on account of using Orca, an Agent IDE, to run 6–10 AI agent sessions in par…
Google rolled the standalone Gemini app with Gemini Live out to Apple CarPlay globally on Wednesday, November…
An OpenAI Japan employee said on November 11 that in-country inference is now available for GPT-6 Luna via the…
Google announced the Gemini Decisions API, built on the Gemini Decision 1 model, which answers up to 1,000 que…
