Google Unveils Decision API; Pricing to Weigh

Chapters
Google launches a narrow question-scoring API
Google announced the Gemini Decisions API in a Blog post on November 11, 2026, describing a developer service powered by a new system called Gemini Decision 1. Google's post said developers can evaluate up to 1,000 yes/no, choice and rubric-scoring questions in a single pass across text, images, audio, video and PDFs. Sundar Pichai announced the launch, Eu 36kr reported on November 12, 2026, noting that a long-awaited Gemini 4 Hydrogen release did not arrive.
A single request can pair those questions with one shared multimodal input under a 256K-token ceiling, according to Google's announcement. Google said the service is built for bounded tasks such as selecting tools, checking policy compliance, routing queries and grading context, Interestingengineering reported on November 11, 2026. Developers could start building on launch day with an API key from Google AI Studio, using guides for Python, JavaScript and REST, Google wrote.
The short version
Google unveils the Gemini Decisions API, a developer service built on a new system called Gemini Decision 1 that scores up to 1,000 yes/no, multiple-choice and rubric questions per request, the company said on November 11. SiliconANGLE and FourWeekMBA reported that TypeSafe AI, OpenAI and Microsoft already sell similar narrow decision services. For a business weighing AI costs, Google said output is free, and ETCIO reported that shared input is billed once per request.
- Eu 36kr put the input price at 1/5 of Flash-Lite's cost.
- Interesting Engineering reported Google's benchmark lead as a claim from the company's own suite.
- ETCIO listed Gemini Notebook queries and Google Search safety as early uses.
- Microsoft's competing Decision-1 opened 33 days earlier, by FourWeekMBA's count.
Google bills shared input once and output at zero
Google lists the Decisions API at $0.10 per million input tokens, with output not billed, according to its November 11 post. The company said the shared input is billed only once, regardless of the number of questions evaluated in a request, Cio Economictimes Indiatimes (ETCIO) reported on November 12, 2026. In its November 12 report, Eu 36kr put the Decisions API input price at 1/5 of the cost of Flash-Lite.
Gemini Decision 1 handles just three question types (yes-no, multiple choice and ratings on a user-specified scale) and answers them at a significantly lower cost, SiliconANGLE reported on November 12, 2026. Microsoft's rival Decision-1 charges $0.042 per million on the input side, also with free output, FourWeekMBA reported on November 11, 2026. The price multiples in its comparison are its own arithmetic, FourWeekMBA noted.
Google supplied the speed and accuracy figures
Google DeepMind product manager Min Choi and principal scientist Felix Yu wrote that Gemini Decision 1 "scores options without token-by-token decoding," SiliconANGLE reported. Gemini Decision 1's median compute latency is 50-100 ms and its end-to-end latency is 150-250 ms on multimodal contexts of up to 16K-token length, staying nearly flat from 1 to 100 questions per request, according to Google's November 11 post.
On benchmarks spanning agent routing, reasoning, ranking and multimodal understanding, Google wrote that Gemini Decision 1 "leads across the suite, with the lowest calibration error." Google benchmarked the new system against Flash-Lite running through the Interactions API, Interesting Engineering reported. Eu 36kr wrote that low calibration error means the confidence Gemini Decision 1 reports matches its actual accuracy. Each answer comes with a confidence score that helps applications decide whether to use the output, SiliconANGLE reported.
TypeSafe AI, OpenAI and Microsoft moved earlier
Jev, launched in mid-September by startup TypeSafe AI Inc., was the first decision model and gained a million users within a few days, SiliconANGLE's November 12 report said. Google compared Gemini Decision 1 with Jev and with OpenAI Group PBC's Decisions API, a competing service that launched last month, according to SiliconANGLE.
Microsoft's Decision-1 opened 33 days before Google's service, FourWeekMBA calculated. Tracker accounts on X listed gemini-decision-1 before Google had posted anything about it, FourWeekMBA reported. Decision services answer their narrow question set faster than LLMs and are easier to build into applications, SiliconANGLE reported.
Early uses cover Search safety and notebook queries
Initial applications include managing queries in Gemini Notebook and boosting safety in Google Search, ETCIO reported. Interesting Engineering's November 11 report said developers can submit yes-or-no checks, multiple-choice questions with up to 255 options, or rubric-based scoring requests in a single API call. A cybersecurity team, for example, could ask Gemini Decision 1 to rate the severity of an antivirus alert on a scale of 1 to 10, SiliconANGLE wrote.
Google's post gave combined-input examples, such as pairing a call recording with account metadata or a scanned PDF claim with damage photos. Developers can also tune resolution on image and video inputs to control visual detail and cost, according to Google.
Tron's take
My take: Google's news is a developer tool, and most owners will meet it through the software they already buy rather than directly. The reported details I would weigh are concrete: output is free, shared input is billed once per request, and the speed and calibration figures come from Google's own benchmark suite. Until independent testers repeat those numbers, I would treat them as vendor claims, the same way XL.net has treated rollout claims from other vendors; that is my reading of the news, not a reported result.
I also read the pace as a reason not to rush. TypeSafe AI, OpenAI, Microsoft and Google all put decision-making APIs on the market between mid-September and November 11, by SiliconANGLE's and FourWeekMBA's accounts. The argument that every small firm must adopt each release immediately does not hold up for me here: a business that waits a quarter can pick the vendor whose prices and accuracy held. The opposite view, that AI news is irrelevant to small firms, misses too, because cheap per-question scoring fits chores like ticket routing and alert ranking, part of the shift toward cheaper, smarter systems.
My advice: ask each software vendor which decision API it calls and what confidence score sends an answer to a person. If a team wants a pilot, SiliconANGLE's example of rating antivirus alert severity is a sensible size, with a human reviewing every low-confidence answer. XL.net sells managed IT and security services, so I have a commercial interest in that kind of work.
Questions I'd expect
What does the Gemini Decisions API do?
It scores closed-ended questions, including yes/no, multiple-choice and rubric ratings, against a shared input of text, images, audio, video or PDFs, and returns probabilities with confidence scores, according to Google and ETCIO.
How is the Gemini Decisions API priced?
Google charges for input only and bills shared input once per request, according to Google and ETCIO. FourWeekMBA reported that Microsoft's rival Decision-1 also leaves output free.
Which companies sell competing decision APIs?
SiliconANGLE named TypeSafe AI's Jev and OpenAI Group PBC's Decisions API as earlier entrants, and FourWeekMBA reported on Microsoft's Decision-1.
Who produced the benchmark results for Gemini Decision 1?
Google published the benchmark results in its own announcement, and Interesting Engineering reported the outperformance as Google's claim.