Google has announced Gemini Decisions API, offering developers a way to handle routine decisions in AI applications without generating text responses, weeks after OpenAI launched a competing service.
Powered by the company’s new Gemini Decision 1 model, the API allows developers to evaluate up to 1,000 questions in a single request. Unlike conventional generative AI models that produce responses token by token, the system assigns probabilities to specified options, allowing applications to make decisions without waiting for a generated answer.
Developers can use the API to check whether a condition is met or select an option from a predefined set. It can also score inputs against a rubric, allowing applications to evaluate the degree to which an input meets specified criteria.
The API supports multimodal inputs, including images, audio, and video, with a context window of up to 256,000 tokens. Developers can use the resulting confidence scores to determine when an application should proceed automatically or refer uncertain cases for further review.
Google said the API delivers end-to-end latency of 150 to 250 milliseconds in tests involving contexts of up to 16,000 tokens, with performance remaining relatively stable across one to 100 questions.
The company has priced the API at $0.10 per million input tokens, without additional output-token charges. “Your shared input is billed once per request no matter how many questions you evaluate (each question adds only its own instruction and criteria tokens),” Google said in its announcement.
Google is already using the technology in its Gemini Notebook product to route queries to different AI models based on the complexity of each request. According to the company, the approach lowered latency for everyday queries by as much as 40%.
Google said it is also using the API in Search to detect adversarial behavior involving Model Context Protocol (MCP) tools, extending its use beyond model routing to security-related decisions.
The API is available through Google AI Studio, while integrations with developer frameworks are being rolled out. Support for Gemini on Google Cloud is expected later, although Google has not specified a release date.
Applications and competition
Specialized decision models could find applications beyond routing requests between AI systems.
“The model is great at generating decisions from unstructured, dynamic input,” said Lian Jye Su, chief analyst at Omdia. He identified policy and risk assessment as potential applications, along with credit scoring.
OpenAI’s Decisions API, which became generally available in October, supports up to 400 questions per request, compared with Google’s limit of 1,000.
Both companies charge $0.10 per million input tokens at standard rates, with no output-token charges. OpenAI also offers discounted pricing for cached inputs, which could affect costs for applications that repeatedly process the same context.
Google’s offering supports a broader range of multimodal inputs than OpenAI’s Decisions API, including audio and video.
Su said a direct comparison between the two products may be less important than how well each fits an enterprise’s existing development environment. He argued that developers should look beyond published benchmarks and assess how reliably these systems make decisions under actual operating conditions.
“The true benefit lies in the reduced misjudgment and improved decision-making quality,” Su said.
Risks in production
Changes in incoming data or deliberately manipulated inputs could cause decision-making APIs to produce incorrect judgments, Su warned. Even when an API returns a confidence score, that alone does not establish whether its assessment is trustworthy.
“All decision-making processes must be highly contextual with support from a trusted and secure context layer,” Su said.
He recommended connecting decision models to databases containing historical transactions and established business policies. This would help ensure that the models evaluate requests against relevant organizational information rather than relying entirely on the immediate input.
Su also said Google would need to meet enterprise requirements for data retention and sovereignty without undermining the cost and performance advantages of the service.
Tulika Sheel, senior vice president at Kadence International, raised a related concern about human accountability as enterprises incorporate AI decision-making into autonomous workflows. “As AI agents become more autonomous, enterprises will need clear accountability for decisions and actions taken on behalf of users,” Sheel said.
Read the original article:
Google’s Gemini Decisions API to handle up to 1,000 questions per request