Is Gemini 3.8 Flash Free for Coding? Antigravity's Free Tier Explained
Google's Antigravity offers free weekly access to Gemini 3.8 Flash for coding agents. Here's how the setup works and what to watch for.
Can you use Gemini 3.8 Flash for free?
Yes, within limits. Google currently includes Gemini 3.8 Flash in Antigravity’s free plan, available through a weekly usage allowance rather than an unlimited subscription. You connect to it through Antigravity’s official ACP (Agent Communication Protocol) server, sign in with a Google login, and point a compatible coding tool at that server. Once connected, Flash shows up as a selectable model for coding tasks, no separate paid API key required, as long as you stay inside whatever quota Google sets for that week.
TL;DR
- Gemini 3.8 Flash is accessible for free through Antigravity’s free plan, which runs on a weekly usage allowance rather than a flat unlimited tier.
- You need the official Antigravity ACP server, not just the regular Antigravity terminal command, to connect a third-party coding tool to Flash.
- Setup happens through provider connections in your coding harness: add the Antigravity agent, point it at the ACP server executable, sign in with Google, then refresh to confirm the connection.
- Free usage is shared across workers, so if you run multiple Flash agents in parallel on the same account, they all draw from the same limit.
- Google’s free allowance can change at any time, so treating it as a permanent unlimited resource is risky for any workflow you depend on.
- Flash works as either a standalone coding model or a cheaper worker alongside a more expensive architect model like Opus, letting you mix cost tiers on one project.
- An API cost estimate is not the same as what you’re billed under the free plan, so don’t assume per-token pricing applies when you’re inside the free quota.
Other agents start typing. Remy starts asking.
Scoping, trade-offs, edge cases — the real work. Before a line of code.
What is Antigravity, and how does it relate to Gemini 3.8 Flash?
Antigravity is Google’s agent framework for exposing Gemini models to coding tools through a standardized connection (the ACP server). Rather than every coding tool needing its own bespoke integration with Google’s models, Antigravity provides a single executable that tools can connect to, handling authentication and model access in one place. Gemini 3.8 Flash is one of the models made available this way, positioned as a lighter, faster option compared to larger frontier models.
For developers, the practical effect is that you don’t need a separate paid subscription to Google’s model just to try it inside a coding agent. You install the ACP server, authenticate with a Google account, and the model becomes available in any tool that supports the connection.
How do you set up free access to Gemini 3.8 Flash?
The setup, as demonstrated in a recent front-end build workflow, follows a specific sequence:
- Install Google’s official Antigravity ACP server. This is a distinct step from installing the regular Antigravity terminal command. The two are not interchangeable, and skipping the ACP server means the connection won’t work even if the terminal tool is installed.
- Sign in through Antigravity’s supported Google login flow. This ties the connection to your Google account and whatever free allowance that account has.
- Open your coding tool’s provider connections settings. Inside the agent or provider configuration menu, add a new connection and point it at the Antigravity ACP server executable on your machine.
- Refresh the agent list to confirm the tool can see the connection. If it doesn’t show up, the executable path or the login step is usually the problem.
- Select Antigravity as the provider, then choose Gemini 3.8 Flash from the list of models your account exposes.
Once connected, Flash behaves like any other selectable coding model: you can hand it a project brief, reference images, design files, and let it write or edit code directly.
Is the free tier actually usable for real coding work?
For lighter or parallelizable tasks, yes. Flash is practical as either a full standalone agent for a project or as a cheaper “worker” model sitting underneath a stronger architect model. In a mixed setup, you might keep a more capable (and paid) model handling architecture and high level decisions, while multiple Flash instances handle independent components like a gallery or a form, each pulling from the same shared free weekly allowance.
The catch is that the allowance is shared across every Flash instance running on your account at once. Running two or three Flash workers in parallel burns through the same quota faster than running one. It’s sensible to start with a small number of parallel workers and watch actual usage before scaling up, rather than assuming headroom you haven’t confirmed.
Most coding tools that connect to Antigravity will surface a usage or limits indicator somewhere near account or agent settings, wherever the provider chooses to expose that information. Checking it before a long session saves you from hitting a wall mid-task.
What are the limits and tradeoffs of using Flash for free?
The biggest practical limit is that “free” here means “free within a weekly quota,” not “free indefinitely.” Google can adjust that allowance, and there’s no guarantee the current limits stay fixed. Anyone building a workflow that depends on Flash access every week should treat the quota as a variable, not a constant.
A second tradeoff is continuity between models. If you start a project with one model (say, a paid frontier model) and then switch a task over to Flash through Antigravity, the new session does not automatically inherit the previous conversation or context. Flash needs to be pointed at the existing code and any design or spec files explicitly before it can pick up where the other model left off. Skipping that step means Flash is working blind on a codebase it’s never seen.
A third point worth flagging: an API-based cost estimate for a model is not the same thing as what you’re charged under a free plan. If you’re used to estimating costs by token pricing, that math doesn’t carry over cleanly to a quota-based free tier. The two systems are billed differently, and conflating them leads to wrong assumptions about what a session actually “costs” you.
Is Gemini 3.8 Flash worth using over a paid model?
It depends on the task. Flash is a smaller, faster model, which tends to make it better suited to narrower, well-specified jobs (a single component, a form, a gallery) than to open-ended architectural decisions across a whole codebase. Paired with a clear brief, reference material, and a design file to keep it consistent, it can handle delegated pieces of a larger project competently while costing nothing extra beyond the time spent inside the weekly quota.
Where it struggles is the same place most smaller models struggle: the less specific the instructions, the more likely the output drifts from what you actually wanted. The free access removes the cost barrier to experimenting, but it doesn’t remove the need for a tight, specific brief. The quality of what you get out of Flash still depends heavily on how much direction you put in, whether that’s written instructions, attached reference images, sketches of layout, or a running design file the model is told to follow.
Frequently Asked Questions
Do I need a paid Antigravity subscription to use Gemini 3.8 Flash?
No. Antigravity’s free plan includes access to Gemini 3.8 Flash within a weekly usage allowance. You do need to install the official ACP server and sign in with a Google account, but no separate paid subscription is required to reach Flash itself.
What’s the difference between the Antigravity terminal command and the ACP server?
The regular Antigravity terminal command is a separate installation from the official ACP server. Connecting a third-party coding tool to Flash requires the ACP server specifically. Installing only the terminal command will not establish that connection.
Can I run multiple Gemini 3.8 Flash agents at once for free?
Remy is new. The platform isn't.
Remy is the latest expression of years of platform work. Not a hastily wrapped LLM.
Yes, but they all draw from the same shared weekly allowance on your account. Running several Flash workers in parallel uses up the quota faster than running one, so it’s worth starting small and checking usage limits before scaling parallel sessions.
Will Gemini 3.8 Flash remember context from a different model’s session?
No. Switching a task from one model to Antigravity’s Flash does not automatically carry over prior conversation history. You need to explicitly point Flash at the existing code and any project or design files so it can pick up the work accurately.
Is Google’s free Gemini 3.8 Flash allowance permanent?
There’s no indication it’s fixed indefinitely. Google can change the limits on its free plan, so it’s best treated as a current allowance rather than a permanent guarantee, especially for workflows that depend on consistent weekly access.
