Google Launches Gemini Decision 1 API for High-Speed Multimodal Scoring
Google DeepMind has introduced the Gemini Decisions API, allowing developers to score up to 1,000 closed-ended questions in a single pass with low latency.
Google DeepMind has entered the emerging space of specialized decision models with the launch of the Gemini Decisions API. Powered by a new model designated as Gemini Decision 1, the interface is designed specifically for fast, closed-ended evaluation tasks, such as binary yes-or-no determinations, multiple-choice selections, and numeric scale ratings. Unlike standard autoregressive language models that generate responses token by token, Gemini Decision 1 evaluates options directly and outputs probabilities alongside confidence scores within a 150 to 250 millisecond window.
Single-Pass Multimodal Processing Across 1,000 Questions
A central feature of Gemini Decision 1 is its ability to process up to 1,000 independent questions simultaneously in a single pass. The system supports a broad range of modalities, accepting inputs across text, images, audio, video, and PDF documents.
Google demonstrated this multi-question capability through a test involving a photograph of a kitchen counter. In a single execution, Gemini Decision 1 evaluated three distinct question formats: a binary stain detection query (returning a confidence value of 0.95), a multiple-choice selection to identify the correct cleaning technique (scoring 98 percent), and a scale-based rating assessing stain severity (yielding 1.19 out of 2).
Before releasing the tool publicly, Google integrated the system into several internal applications. The company currently utilizes the Gemini Decisions API for dynamic model routing within Gemini Notebook, rapid design prototyping in Stitch, and automated safety verification systems in Google Search.
Benchmark Performance and Cost Structure
In standard evaluation suites, Gemini Decision 1 leads performance benchmarks in categories such as agent routing, logical reasoning, and multimodal comprehension. Google also highlighted that the model records the lowest Top-1 Expected Calibration Error (Top-1 ECE), reflecting high accuracy in its output probability distributions.
To accommodate high-volume enterprise workloads, Google priced the API at $0.10 per million input tokens, with no charges applied for output tokens. According to Google, this pricing structure renders Gemini Decision 1 approximately five times cheaper to operate than Gemini 3.5 Flash-Lite.
The Gemini Decisions API is available immediately through Google AI Studio. Integrations are also currently rolling out across third-party platforms and developer frameworks, including OpenRouter, LangChain, and Vercel. For engineering teams looking to integrate diverse model architectures into their workflows, platforms like https://apixoai.online enable developers to try top AI models cheaply through one API key.
What it means for developers
The introduction of Gemini Decision 1 gives software engineers a streamlined alternative to using full-scale generative language models for simple evaluation and routing logic. In traditional architectures, sending binary choices or classification tasks to large language models often incurs unnecessary latency and output token costs. By providing direct confidence scores in under 250 milliseconds, specialized decision models allow developers to build faster guardrails, content moderation tools, and decision-tree logic.
Additionally, the capacity to process 1,000 questions in a single pass significantly reduces architectural complexity. Developers no longer need to construct multi-step prompt chains or orchestrate parallel API calls to analyze different attributes of a single document or media file.
Google's release comes amid growing industry interest in specialized decision engines. The trend gained momentum following an initial release by startup Jev, which was rapidly followed by similar product announcements from OpenAI, Cloudflare, and Microsoft. As major providers expand their offerings in this segment, developers gain access to dedicated micro-task models optimized for latency, cost, and strict classification tasks.
Source: Google joins the decision-model wave with Gemini Decision 1, scoring 1,000 questions in a single pass — The Decoder. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
