Antigravity Boost Mode: Pricing, Quota, and Plan Requirements
Boost mode in Google Antigravity needs a paid plan and draws down quota. Here's how it works with Gemini 3.8 Flash and what it costs.
What does Boost mode cost in Google Antigravity?
Boost mode requires a paid Antigravity plan and consumes your plan’s usage quota, with any overages billed according to your account’s credit settings. There is no separate published “Boost multiplier” that tells you exactly how much more a boosted request costs versus a normal one. Google’s own API pricing for the underlying model, Gemini 3.8 Flash, starts at 75 cents per million input tokens and $3.75 per million output tokens as an introductory rate through the end of 2026, but that’s API pricing, not a direct readout of what Boost consumes inside the Antigravity product.
TL;DR
- Boost is a workflow, not a model. You select Gemini 3.8 Flash as your model and trigger Boost with a slash command in the prompt box, which routes the task through a multi-agent process instead of a single response.
- A paid plan is mandatory. Google’s documentation states Boost requires a paid Antigravity plan, so free-tier users can select Flash but can’t invoke the Boost workflow.
- Quota and credits move independently. Antigravity draws down your plan’s remaining quota based on the actual work performed, and any overage is billed according to whatever credit settings are attached to your account.
- Gemini 3.8 Flash has adjustable reasoning effort. Higher effort levels let the model spend more tokens working through a problem, which means a “cheap” model can still rack up real cost once agent work multiplies the number of calls.
- API pricing and in-app cost are two different things. The 75 cents/$3.75 per million token rates are what developers pay through the API; inside Antigravity itself, cost is expressed as quota consumption, not a line-item token bill.
- No fixed cost multiplier has been established for Boost. One creator’s hands-on test explicitly avoided claiming a specific overhead number, since Google hasn’t published one and a single demo can’t reliably measure it.
- Boost is built for hard problems, not everyday edits. It’s aimed at difficult bugs, algorithmic work, and complicated refactors, where the extra coordination overhead is more likely to pay off than on routine coding tasks.
Seven tools to build an app. Or just Remy.
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
How does Boost mode actually work?
Boost is invoked with a slash command (/boost) typed into the prompt box after you’ve selected your model, in this case Gemini 3.8 Flash, underneath the message field. Once triggered, the workflow doesn’t just send your prompt straight to the model. According to Google’s documentation, it first works out a strategy for the task, then breaks that strategy into focused jobs handed off to sub-agents, including agents dedicated to implementation and agents dedicated to investigation.
Those sub-agents check their own results before a coordinating agent merges everything into a final answer, adding a verification pass on top. That structure is why Boost is positioned for tasks like difficult bugs, non-trivial algorithms, and complicated refactors rather than quick edits. The tradeoff is straightforward: more agents doing more checking means more total model calls and more tokens processed to complete the same task, compared to a single-shot response.
Why does Boost require a paid plan?
Because it’s fundamentally more expensive to run. A single Boost request isn’t one inference call, it’s a coordinated sequence of planning, delegated sub-agent work, self-checking, and reassembly. Each of those steps consumes tokens, and Gemini 3.8 Flash’s adjustable reasoning effort means the model can lean even harder into token spend on genuinely difficult problems, particularly at higher effort settings.
Antigravity’s plan requirement is the practical mechanism for controlling that cost. Free access gets you the base model experience; Boost’s multi-agent overhead is gated behind a paid tier so that the heavier compute draw lines up with an account that has quota and billing settings attached to it. This is a common pattern in agentic coding tools: the model itself might be inexpensive per token, but wrapping it in an agentic loop with multiple sub-tasks can multiply the effective cost of a single user request many times over.
How does quota consumption work with Boost?
Antigravity’s documentation ties quota consumption to the actual work performed, not to a flat per-request charge. That means a Boost request that spins up several sub-agents to investigate a bug and verify a fix will draw more from your quota than a simple, low-effort exchange. Credit overages beyond your plan’s quota are governed by whatever credit settings you’ve configured on your account.
This is worth sitting with before you launch a large or open-ended Boost session. Because the workflow is designed to expand into deeper investigation when a problem calls for it, an ambiguous or broad request can trigger more sub-agent work than a tightly scoped one. Checking your remaining quota and credit configuration before a big run is a reasonable habit, since Boost’s whole design premise is that it does more work automatically when a task looks hard.
What is Gemini 3.8 Flash’s actual API pricing?
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
Google’s introductory API pricing for Gemini 3.8 Flash is 75 cents per million input tokens and $3.75 per million output tokens, and that rate is set to hold through the end of 2026. That’s the developer-facing API price for calling the model directly, separate from how Antigravity meters usage internally through plan quota.
The distinction matters because a low per-token API price doesn’t automatically translate into a low total cost for a finished project. If a Boost-driven session calls the model repeatedly through planning, sub-agent execution, and verification, the output-token volume for a single coding task can add up well beyond what a single prompt-and-response exchange would cost at the same per-token rate. Anyone estimating project cost from the headline price alone is likely to undercount.
Is Boost mode worth the added cost?
It depends on the task. Boost is explicitly built for harder problems: difficult bugs, algorithmic logic, and complicated refactors where a deeper, self-checking, multi-agent process is more likely to catch mistakes that a single-pass response would miss. On a real test build, a maze game with procedural generation and an animated pathfinding solver, the extra process actually mattered. An independent review step caught a bug where regenerating a maze could silently drop event subscriptions, plus a movement validation flaw that let a player jump through a wall if the game engine was called directly instead of through the UI. Boost sent those findings back into the workflow, which fixed the engine and added regression tests for both issues, and the resulting suite passed seven tests including 100 generated mazes and a separate check across 600 mazes.
That’s a meaningful result, but it’s a single demonstration, not a controlled benchmark. There’s no established, repeatable measurement showing Boost consistently outperforms plain Gemini 3.8 Flash across a range of tasks, and no fixed multiplier for how much more quota or credit a boosted run consumes versus a standard one. The reasonable takeaway is that Boost is a documented option for paid users who want deeper reasoning on genuinely hard problems, best used with a clear, specific definition of what “working” looks like, rather than a blanket upgrade to reach for on every coding task.
Frequently Asked Questions
Does Boost mode work with models other than Gemini 3.8 Flash?
The demonstrated and documented use case pairs Boost with Gemini 3.8 Flash, letting you keep Flash selected while requesting the deeper multi-agent workflow for harder tasks. The source material doesn’t cover Boost’s availability with other models in Antigravity.
How do I turn on Boost mode in Antigravity?
Select your model underneath the prompt box, then start your message with the /boost command followed by a description of the task. Boost then runs its own strategy, delegation, and verification steps before returning a result.
Is there a free way to use Boost mode?
No. Google’s documentation states Boost requires a paid Antigravity plan. Free-tier access lets you use models like Gemini 3.8 Flash directly, but not the Boost workflow layered on top.
Does a low API price mean Boost is cheap to run?
Remy is new. The platform isn't.
Remy is the latest expression of years of platform work. Not a hastily wrapped LLM.
Not necessarily. The 75-cent input and $3.75 output per-million-token API pricing is the base rate for Gemini 3.8 Flash. Boost’s multi-agent structure, planning, sub-agent execution, self-checking, and reassembly, can involve substantially more total token usage than a single response, and Antigravity meters that as quota consumption rather than a direct per-token bill.
Can I predict exactly how much a Boost session will cost in advance?
Not precisely. Antigravity’s quota consumption scales with the actual work performed, and Google hasn’t published a fixed cost multiplier for Boost versus standard requests. Checking your remaining quota and credit settings before a large or open-ended Boost run is the most practical safeguard currently available.