Google has released the Computer Use tool in its Gemini API to general availability across the Interactions and GenerateContent APIs, according to an entry dated 30 October in the Gemini API release notes. The tool lets developers build agents that control a browser, a phone or a desktop from screenshots.
The first Computer Use entry in those release notes is dated 7 October 2025: “Launched Gemini 2.5 Computer Use Preview”. From that preview to general availability took 388 days, by our arithmetic. The public preview in Gemini 3.5 Flash, dated 24 June 2026, came 128 days before.
Short explainer · Gemini Computer Use goes GA after 388 days
A vertical news Short summarising this story. Figures come from Google’s Gemini API release notes, docs and pricing page; our arithmetic is labelled.
Business Pill · The Model and the Harness
An evergreen Business Engineer explainer on how an AI model and the harness around it split the work. It teaches the general idea only and does not discuss Gemini Computer Use.
The key insight:Google has moved Computer Use to general availability 388 days after its first preview, by our arithmetic, and bills it as ordinary tokens rather than as a separate product.
The Computer Use documentation says the tool lets the model “see” a computer screen through screenshots and “act” by generating UI actions such as mouse clicks and keyboard inputs. The developer runs the actions on their own side, in a loop: send a screenshot, receive an action, run it, send the new screenshot.
The page lists four models that can use the tool: Gemini 4 Argon, Gemini 3.8 Flash, Gemini 3.5 Flash-Lite and Gemini 3 Flash Preview. It calls Gemini 3.8 Flash “The recommended model for computer use”.
For Gemini 4 and Gemini 3.x models, the page says each action comes with an “intent” field explaining the model’s reasoning. It lists browser, mobile and desktop environments as supported.
Computer Use entries in the Gemini API release notes, from the Gemini 2.5 Computer Use Preview (7 Oct 2025) to general availability (30 Oct 2026), and Gemini 3.8 Flash token prices before and after 1 Jan 2027. Day counts are FourWeekMBA arithmetic.
The Safety Rules Around Checkout
The documentation says a response may include a “safety_decision” from an internal safety system. That system classifies an action as allowed, “require_confirmation” or blocked, and the developer’s code is expected to halt on a block.
The page lists seven safety policy categories, by our count. The first, FINANCIAL_TRANSACTIONS, “Blocks or triggers confirmation for actions involving payments, retail checkout, or regulated goods”. Others cover sending emails or messages, creating accounts, and accepting terms of service.
Developers can override select policies, the page says, but “the model may still return a safety_decision with require_confirmation in some cases”. Prompt injection detection, which scans screenshots for hidden instructions, is opt-in: “The default is false”.
The page also carries a warning: “Computer Use may contain errors and security vulnerabilities”. It recommends close supervision for important tasks.
The Business Lever: Billed as Tokens
Google does not charge a separate fee for the tool. The pricing page lists computer use as “Charged as regular tokens per modelpricing” on the paid tier and “Not available” on the free tier.
The pricing page lists Gemini 3.8 Flash, the recommended model, at $0.75 per million input tokens and $3.75 per million output tokens through 31 December 2026. From 1 January 2027 it lists $1.50 and $7.50.
By our arithmetic, that is a doubling of both rates 63 days after general availability. Gemini 4 Argon, the most capable listed model, is priced at $2.00 per million input tokens through 30 November, rising to $4.00 from 1 December, according to the same page.
The Structural Read
As we read the documentation, the tool hands Google’s models the same job as an agent harness: look at a screen, choose an action, and wait for the developer’s code to run it.
As we read it, the safety layer sits at the points where money and consent change hands: payments and checkout, account creation, sending messages and accepting terms.
As we read the pricing page, the cost of running these agents is tied to the model’s token rates, and the recommended model’s listed rates double from 1 January 2027.
Gemini API docs, Computer use
“Blocks or triggers confirmation for actions involving payments, retail checkout, or regulated goods.”
Three Implications
FOR DEVELOPERS The documentation says developers must run the actions on their own side and recommends a sandboxed VM or container.
FOR MERCHANTS The FINANCIAL_TRANSACTIONS category covers payments and retail checkout, according to the documentation.
FOR GOOGLE The pricing page lists computer use as charged at regular token rates on the paid tier and not available on the free tier.
The framework says “Orchestration, memory, tool use, and workflow execution have become the new sources of competitive advantage.”
As we read it, Computer Use is Google packaging tool use and workflow execution into its API at token prices. The framework is the owner’s argument and does not discuss Gemini Computer Use or its pricing.
The pages read do not give usage figures, customer names, success rates or benchmark scores for Computer Use. They do not say how many developers used the previews, or what a typical agent task costs in tokens.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Google’s Gemini API release notes list Computer Use as generally available from 30 October, 388 days after the first Computer Use preview. The tool is billed at regular token rates, and the pricing page lists the recommended model’s rates at double from 1 January 2027.
A note on sourcing. All facts and figures come from Google’s Gemini API release notes, Computer Use documentation and pricing page, read on 31 October 2026. Day counts and the price comparison are our arithmetic. We did not contact Google.
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.
