Google Launches Gemini Decisions API With 1,000-Question Limit as OpenAI Caps at 400
Updated
Updated · InfoWorld · Nov 12
Google Launches Gemini Decisions API With 1,000-Question Limit as OpenAI Caps at 400
3 articles · Updated · InfoWorld · Nov 12
Summary
Google said its new Gemini Decisions API can evaluate up to 1,000 questions in one request, letting apps make routine choices by scoring predefined options instead of generating text.
The service uses Gemini Decision 1 to return probabilities and confidence scores across text, images, audio and video, with a 256,000-token context window and 150-250 millisecond latency in tests up to 16,000 tokens.
Pricing starts at $0.10 per million input tokens with no output-token charges, and Google says shared input is billed once per request regardless of how many questions are evaluated.
Google is already using the technology in Gemini Notebook, where it cut latency on everyday queries by as much as 40%, and in Search to detect adversarial behavior involving MCP tools.
The launch comes weeks after OpenAI rolled out a rival Decisions API with a 400-question cap; analysts said real-world reliability, governance and human accountability will matter more than headline benchmarks.
Will Google's lack of input caching make the Gemini Decisions API a hidden financial trap for high-volume enterprise users?
How does Gemini's decision model weigh conflicting signals when analyzing multimodal inputs like video and text simultaneously?
Could delegating routine enterprise choices to probabilistic AI models create unforeseen security vulnerabilities through adversarial input manipulation?