Sundar Pichai wrote on X on 11 November that Google’s new Gemini Decision 1 costs “$0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite)”. Google’s own benchmark table supports the figure on its own test suite: the seven shared evaluations cost $0.88 to run on Gemini Decision 1 and $4.14 on Gemini 3.5 Flash-Lite, 4.7 times as much, by our arithmetic.
On Google’s list prices the gap is not fixed. In five worked cases built on Google’s pricing page and our own token assumptions, Flash-Lite costs between 3.25 and 28 times as much per 1,000 decisions in four cases, and about a third as much in the fifth, by our arithmetic.
Business Pill · Tokens Explained
An evergreen explainer on tokens, the unit AI models read, write and bill in. It teaches the general idea and does not discuss Gemini Decision or Flash-Lite.
The key insight: As we read Google’s pages, the 5x is a property of the workload, not of the price list: Gemini Decision charges a third of Flash-Lite’s input price and nothing for output, so its edge grows with every token Flash-Lite writes and falls to about 3.25x in our case where Flash-Lite returns only a 5-token label. Where a document must be re-sent with each question, Flash-Lite’s caching can reverse it.
Google’s pricing page lists Gemini Decision at $0.10 per million input tokens, with output “Not applicable (0 output tokens)”. It lists Gemini 3.5 Flash-Lite at $0.30 per million input tokens and $2.50 per million output tokens, a line it labels “Output price (including thinking tokens)”.
On input alone, Flash-Lite costs three times as much, by our arithmetic. Any gap above 3x comes from the tokens Flash-Lite writes back, which Gemini Decision does not produce: Google’s Decisions API guide says “It generates zero text tokens, so output token usage is always 0”.
Google’s launch post gives the 5x as a benchmark result, not a list-price ratio: Gemini Decision 1 “is nearly 5x cheaper across the shared benchmark suite than Gemini 3.5 Flash-Lite with the Interactions API”. The table labels that Flash-Lite column “Interactions API (minimal)”. The table and its costs are Google’s own; we did not run the evaluations.
Cost per 1,000 decisions on Gemini Decision and Gemini 3.5 Flash-Lite in five worked cases, at the list prices on Google’s Gemini API pricing page (11 November 2026). Token counts per case are our assumptions; costs and multiples are our arithmetic. Case E assumes Flash-Lite caches a 10,000-token document for one hour.
What Was Posted on X
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio.
Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details:
– 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring – $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite) – leads on several benchmark evals with lowest calibration error
Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
Sundar Pichai’s launch post at 18:15 UTC lists the three decision types, the $0.10 per 1M input token price and the ‘~5x cheaper than Flash-Lite’ claim that this piece checks against Google’s pricing page and benchmark table.
Five Worked Cases
Our cases use Google’s list prices and token counts we chose; they are assumptions, not measurements. Case A is 1,000 separate requests of 500 input tokens each, with Flash-Lite answering each in a 5-token label. Gemini Decision costs $0.05 and Flash-Lite about $0.16, or 3.25 times as much, by our arithmetic.
Case B lets Flash-Lite add a short reason, 40 output tokens per answer: it costs $0.25, five times Decision’s $0.05. Case C counts 500 output tokens per answer, including thinking tokens: Flash-Lite costs $1.40, or 28 times as much, by our arithmetic.
For one question per request without caching, Flash-Lite’s cost is three times Decision’s plus 25 times the ratio of its output tokens to input tokens, by our arithmetic. That reaches 5x when Flash-Lite writes 8% as many tokens as it reads, such as 40 tokens back on a 500-token input.
Case D puts one 10,000-token document and 1,000 questions of 20 tokens each into a single request, and has Flash-Lite answer all of them in one prompt at 5 tokens each. Decision costs $0.003 and Flash-Lite $0.0215, about 7.2 times as much, by our arithmetic.
Case E sends the same document with each question in its own request. The guide says the Decisions API does not support context caching or the Batch API, and that “Repeated input across requests counts in full”. Decision then costs about $1.00. Flash-Lite, with the document cached for one hour at the page’s $0.03 per million cached tokens and $1.00 per million tokens per hour of storage, costs about $0.33. Decision costs about three times as much in that case, by our arithmetic.
What the Single Pass Changes
Google’s post says the model “scores options without token-by-token decoding”, with latency “remaining nearly flat from 1 to 100 questions per request”. The pricing page says “Each request bills shared input tokens once, regardless of question count”, and that each question adds tokens for its own instructions, criteria and formatting.
That billing rule moves the cost more than the choice of model in our cases. The same 1,000 decisions on Gemini Decision cost $0.003 packed into one request (Case D) and $1.00 sent one at a time (Case E), 334 times as much, by our arithmetic.
Google’s launch post says a multiple-choice question “Selects from 2 to 255 choices, returning the winner and normalized probabilities for every candidate”. The guide lists 2 to 255 options per choice question, and the pricing page bills each question’s criteria as input: 255 options of about five tokens each add about 1,275 tokens, or about $0.13 per 1,000 such questions, by our arithmetic. A text model would read the same options and write back only its pick.
Google’s page still lists the older Gemini 2.5 Flash-Lite at $0.10 per million input tokens for text, image and video, and $0.40 for output. In Case A it would cost $0.052, against $0.05 for Gemini Decision, by our arithmetic. Google’s benchmark table does not include that model.
What Google’s Table Adds
Pichai’s post also says the model “leads on several benchmark evals with lowest calibration error”. Google’s table gives a top-1 calibration error of 0.0250 for Gemini Decision 1 and 0.2104 for Flash-Lite across seven shared evaluations, where lower is better. These are Google’s self-reported figures.
Our launch piece compares the price with Microsoft’s Decision-1, and our earlier piece covers the first sighting of the model. Our piece on Amazon’s label model covers a similar move from a general model to a narrow one.
The Structural Read
As we read the billing rule, the bigger saving in our cases comes from packing questions into one request: the same 1,000 decisions on Gemini Decision cost $0.003 in one request and $1.00 sent one at a time, by our arithmetic.
As we read Google’s table, its own 4.7x figure, $0.88 against $4.14 by our arithmetic, sits between our label-only and short-reason cases; the pages we read do not give the token counts behind it.
Google, Gemini Decisions API launch post, 11 November 2026
“There are no output-token charges, variable thinking-token bills, or separate audio transcription and OCR costs.”
Three Implications
FOR DEVELOPERSGoogle’s guide says the Decisions API does not support context caching or the Batch API, and that repeated input across requests counts in full.
FOR COST MODELSGoogle’s pricing page lists Gemini 3.5 Flash-Lite output, including thinking tokens, at $2.50 per million, against $0 for Gemini Decision.
FOR BUYERS OF THE CLAIMGoogle’s blog states the 5x as cost across its shared benchmark suite; its table puts that cost at $0.88 for Gemini Decision 1 and $4.14 for Flash-Lite.
The framework records: “The new job of finance is to manage the ratio between the two — work produced per token burned. That ratio is the gross-margin equation of the agentic era.”.
As we read it, a decision model changes the denominator for a narrow class of work: Google bills Gemini Decision on input only, so the tokens a text model would spend writing an answer drop out of the bill. The framework predates the launch and does not discuss Gemini Decision.
What Is Not Established
Every token count in our cases is an assumption; real counts depend on the prompt, the tokenizer and the media. The pages we read do not say how many thinking tokens Flash-Lite used at the ‘minimal’ setting in Google’s table, so we cannot say which of our cases is closest to Google’s test.
We did not model batch pricing: the pricing page lists “Batch API (50% cost reduction)” on its paid tier, and the guide says the Decisions API does not support it. We did not model caching minimums or terms beyond the listed prices, and we did not contact Google for this piece.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Pichai’s “~5x cheaper than Flash-Lite” matches Google’s own benchmark table, $0.88 against $4.14, or 4.7x by our arithmetic. On list prices the floor is 3x on input alone, and the rest depends on how much Flash-Lite writes: 5x at 40 output tokens on a 500-token input, 28x with 500 thinking and answer tokens, and a reversal when Flash-Lite can cache a document that Gemini Decision must re-read in full.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. We read Google’s blog post, the Gemini API pricing page and the Decisions API guide on 11 November 2026, re-reading them at about 19:25 UTC, and Sundar Pichai’s post on X. The benchmark and cost figures are from Google’s own table and are self-reported. The token counts in the five cases are our assumptions; costs, multiples and the break-even rule are our arithmetic on Google’s list prices. We did not contact Google for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
