Editorial illustration for Gemini Decisions API cuts costs nearly 5x by billing zero output tokens
Gemini Decisions API Cuts Costs 5x With Zero Output Tokens
Google launched the Gemini Decisions API on Tuesday, built around a new model called Gemini Decision 1, available through Google AI Studio. The pitch is narrow by design: instead of generating full text responses, the model scores a fixed set of candidate options in one forward pass and hands back probabilities and confidence scores. That's useful for things like picking which tool an agent should call, checking a response against a policy, routing a query to the right model, or grading whether retrieved context actually answers a question.
The pricing structure is what separates this from a standard chat endpoint. Gemini Decision 1 runs at $0.10 per 1 million input tokens, and because it never decodes an answer token by token, there's no output to bill for. A single request can carry multimodal input, text, images, audio, video, or PDF, up to 256,000 tokens, bundled with as many as 1,000 independent questions answered in one pass. Google says the combination of skipping generation and batching questions together is what drives the latency and cost numbers down, and the company is framing the results against its own Gemini 3.5 Flash-Lite running through the standard Interactions API.
You can pass one multimodal input across text, images, audio, video, and PDFs (up to 256K tokens) along with up to 1,000 independent questions. Rather than generating answer text token by token, Gemini Decision 1 scores your candidate options in a single forward pass, returning probabilities and confidence scores in 150-250 ms end-to-end latency at $0.10 per 1M input tokens with zero output tokens billed.
Why this matters
For teams running high-volume classification work (tool routing, policy checks, content moderation) token billing has always been the hidden tax on top of latency. A 5x cost cut on a shared benchmark suite, achieved simply by not billing output tokens, is the kind of change that reshapes which products are economically viable at scale rather than just a nice-to-have speed bump. If you're currently forcing a general-purpose model like Gemini 3.5 Flash-Lite through the Interactions API to make a yes/no or pick-one-of-N decision, you're paying for generation you don't need.
Gemini Decision 1 bets that closed-set decisions are common enough, and distinct enough from open-ended generation, to deserve their own primitive and pricing model. We'd want to see independent benchmarks before taking "nearly 5x cheaper" at face value, since Google is comparing against its own product line here. But the architectural idea, treating decisions as a separate class of inference from generation, is worth watching.
If other providers follow with their own decision-specific endpoints, the economics of agentic tool selection and routing could shift meaningfully across the industry, not just inside Google's stack.
Related Reading
Browse all Google news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google to End Free Access to Gemini AI for Non-Subscribers
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic NVIDIA and Google Cloud let developers scale AI from prototype to production
Market Trends
Related Topic OpenAI raises bug bounty payouts to USD 172,000, matching Google rates
Business & Startups
From Archives OpenAI invites brands to set up shop in ChatGPT amid AI Futures Project research
From Archives MBZUAI IFM Releases PAN 2, an Open 29B World Model for Physical AI
From Archives Google launches Gemini 4 Carbon, claiming cybersecurity gains over rivals
Business & Startups
From Archives DeepMind Reveals Gemini Robotics 2.5 and Open Safety Benchmark ASIMOV-Jailbreak
Research & Benchmarks
Common Questions Answered
How does Gemini Decisions API reduce costs compared to traditional text generation models?
Gemini Decisions API cuts costs by approximately 5x by billing only input tokens and charging zero output tokens. Instead of generating full text responses token by token, the model scores a fixed set of candidate options in a single forward pass and returns probabilities and confidence scores, eliminating the output token billing that typically adds significant expense to high-volume classification tasks.
What types of multimodal inputs does Gemini Decision 1 support?
Gemini Decision 1 supports multiple input formats including text, images, audio, video, and PDFs, with a combined capacity of up to 256K tokens. Users can pass one multimodal input along with up to 1,000 independent questions in a single request, enabling flexible and comprehensive decision-making scenarios.
What are the practical use cases for the Gemini Decisions API?
The Gemini Decisions API is designed for tasks such as selecting which tool an agent should call, checking responses against policies, routing queries to the appropriate model, and grading content quality. It's particularly valuable for high-volume classification work like tool routing, policy checks, and content moderation where cost efficiency at scale is critical.
What is the latency and pricing structure of Gemini Decision 1?
Gemini Decision 1 delivers end-to-end latency of 150-250 milliseconds and charges $0.10 per 1 million input tokens with zero output tokens billed. This pricing model makes it economically viable for teams running high-volume decision-making operations that would be prohibitively expensive with traditional text generation models.
Further Reading
- Gemini Decisions API: Fast, calibrated multimodal decisions - Google
- Gemini Decision 1 API, Pricing & Playground - Vercel AI Gateway
- Gemini Developer API pricing - Google AI for Developers
