Google has moved Antigravity into the Gemini API as a managed coding agent, and the preview pricing gives developers a real reason to test it now rather than wait.
Until now, Antigravity was a standalone IDE. It was a place developers opened when they wanted an AI agent to write code and run commands - working through a task with less hand-holding. That changes this month. Google's developer docs, last updated on September 2, 2026, now list Antigravity as a managed agent inside the Gemini API and Google AI Studio, available through the Interactions API for both free tier and paid tier projects.
The hook is the packaging. One request can provision a Google-hosted Linux sandbox where the agent reasons, executes code, manages files and browses the web. You don't have to wire together a model call, a shell, a search tool and a file system just to see whether the thing can finish a job. Google is trying to make the agent itself the product surface.
It runs on Gemini 3.8 Flash by default, with options to configure earlier Gemini Flash models through agent settings. According to Google's Antigravity documentation, the agent uses the same harness as the Antigravity IDE and supports code execution, Google Search, URL fetching and file handling through the remote environment. The docs also describe the loop plainly: the agent plans, acts, observes results and keeps going until it finishes or hits a limit.
That last part matters if you build with it. A normal chat request gives you an answer. An Antigravity interaction can burn through tokens while it writes a script, runs tests, reads the output, changes the file and tries again. Google's own docs warn that complex workflows can reach 3 million to 5 million tokens in a single interaction, with estimated costs of up to about $5 depending on the task.
Google Ships Gemini 3.8 Flash, Its Third Flash Model in Six Weeks Google released Gemini 3.8 Flash on September 2, its third Flash-tier model in six weeks, undercutting rivals on price while beating GPT-5.6 Terra and Claude Sonnet 5 on the Terminal-Bench 2.1 coding benchmark. The rapid cadence signals Google is racing to win on cheap, fast models rather than just the biggest ones. - google gemini 3.8 flash model pricing per token - cheap fast ai model releases six weeks cadence
The cheap part is the opening move #
Google Cloud's pricing page lists Gemini 3.8 Flash, 3.7 Flash and 3.6 Flash at introductory pricing of $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. On January 1, 2027, that standard pricing is scheduled to double to $1.50 for input and $7.50 for output. That's still cheap enough for serious testing, but it's not the same as free.
The free part is narrower and more interesting. Google says free tier projects include a free rate limit and usage quota, and its Antigravity docs say environment compute - CPU, memory, plus sandbox execution - is not billed during the preview period. In plain English, you may still pay for model tokens and tools, but Google isn't charging separately for the Linux machine doing the work right now.
Frankly, that's how you get developers to try an agent that might otherwise feel risky. If you're testing a workflow that installs packages, runs a build, fixes a failing test and writes a short report, the sandbox meter is one more reason to hesitate. Google has removed that reason for the preview. It hasn't removed the need to watch token use, which is why the docs push budget controls such as max_total_tokens.
OpenAI and Anthropic are the real comparison #
The Interactions API is not a side door. Google announced on June 22, 2026, that it had reached general availability and had become the company's primary interface for Gemini models and agents. The old generateContent API remains supported, but Google's direction is clear: one interface for model calls, background work, managed agents and long-running tasks.
That puts Antigravity directly beside OpenAI and Anthropic's agent tools. OpenAI introduced AgentKit in October 2025, but later said on its own site that Agent Builder and Evals would no longer be available on the OpenAI platform after November 30, 2026. It pointed developers toward the Agents SDK for workflows that should continue as code. That's not a small distinction. Google is putting a managed coding agent into its main API just as OpenAI is pruning part of the visual builder story around AgentKit.
Anthropic has taken a more code-first route with the Claude Agent SDK. Its documentation focuses on permissions, tools, hooks and modes such as acceptEdits, dontAsk, plan and bypassPermissions. You define what Claude is allowed to do: read files, edit code, run commands. Then you let the agent work inside those boundaries. Google's Antigravity preview looks closer to that world than to a drag-and-drop builder, except Google is hosting the sandbox for you.
For developers, the choice is practical. If you already live inside Gemini, Antigravity now gives you a ready-made way to test agentic coding without building the scaffolding first. If you're already using Claude Code or OpenAI's Agents SDK, Google still has to prove that Antigravity is better than what you've wired into your work. Price helps. It doesn't settle the argument.
[Superhuman Buys Fathom, Betting Notetakers Belong Inside the Inbox](https://startupfortune.com/superhuman-buys-fathom-betting-notetakers-belong-inside-the-inbox/)
Superhuman, the AI productivity company formerly known as Grammarly, has acquired YC-backed meeting notetaker Fathom rather than finish building a rival in-house. Fathom, valued at $94 million in 2024, brings more than 400,000 monthly users and backers like Reddit CEO Steve Huffman into Superhuman's growing suite of email, docs and agent tools. - Superhuman acquisition of Fathom meeting notetaker app - AI notetaking tool integrated into email inbox
The preview works now. It's useful, and temporary by design. Google's docs warn that Antigravity's features and schemas may change, which is exactly what you should expect from a managed agent still carrying a preview label. Build experiments on it now. Don't pretend the terms are fixed.
Also read: Superhuman Buys Fathom, Betting Notetakers Belong Inside the Inbox • Micron stock drops 5% as Amodei, Altman and Musk warn AI is moving too fast • How Does AI Model Distillation Work for Startups Right Now
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.