Antigravity & Gemini 4 Argon guide

Which models can I use?

Access depends on your plan and the Antigravity client. A paid Pro subscription unlocks Argon and Claude 5.5; a Pro trial does not.

Scroll to compare plans. Model names stay in view; filtering a plan shows only its available models.

Antigravity model access, reviewed 18 October 2026; subject to quota and rollout
Reasoning model	Free / Plus	Pro trial	Paid Pro	Ultra	Enterprise
Gemini 4 Argon Argon specs & API prices	Not included	Not included	Included	Included	Included
Gemini 3.8 Flash	Included	Included	Included	Included	Included
Gemini 3.7 Flash	Not included	Not included	Not included	Not included	Included
Gemini 3.6 Flash	Not included	Not included	Not included	Not included	Included
Gemini 3.1 Pro	Not included	Not included	Not included	Not included	Included
Claude Sonnet 5.5 (thinking)	Not included	Not included	Included	Included	Admin opt-in
Claude Opus 5.5 (thinking)	Not included	Not included	Included	Included	Admin opt-in
Claude Sonnet 4.6 (thinking) Removal: 2 Nov 2026	Included	Included	Included	Not included	Not included
Claude Opus 4.6 (thinking) Removal: 2 Nov 2026	Included	Included	Included	Not included	Not included
GPT-OSS-120B Removal: 2 Nov 2026	Included	Included	Included	Included	Not included

Argon client support: App, CLI, and IDE extensions. It is unavailable in the standalone Antigravity IDE.
Scheduled removal: Claude Sonnet 4.6, Claude Opus 4.6, and GPT-OSS-120B leave Antigravity on 2 November 2026.
Enterprise partner models: Claude requires administrator opt-in and is billed by token consumption.

Sources: model availability, paid-plan eligibility, Enterprise model billing.

Gemini 4 Argon, at a glance

Google's model specifications

Gemini 4 Argon is the official model name. It is available in Antigravity for paid Pro and Ultra subscribers, and through eligible Enterprise setups. Direct Gemini API usage has separate billing.

Direct Gemini API specifications; these do not define an Antigravity subscription's allowance
Property	Gemini 4 Argon
API model ID	`gemini-4-argon`
Inputs / output	Text, images, video, audio, and PDF in; text out.
Shared token window	1,048,576 tokens across input, thinking, and output.
Default output limit	65,536 tokens, including thinking.
Long decode	Beyond 65,536 requires background execution and a higher output limit; the shared window still applies.
Thinking levels	Low, medium (default), and high in the Gemini API.
Caching	Implicit only, from 4,096 prompt tokens; cache hits receive a 95% input discount. Explicit caches are unsupported.

For example, 200,000 input tokens leave 848,576 tokens for thinking and output within the shared window. The default output limit remains lower unless you explicitly configure long decode.

Sources: model limits and supported formats, long decode, thinking, and implicit caching, Antigravity rollout.

What does the Argon API cost?

Google's API pricing

These are Gemini Developer API token rates. They are not Antigravity subscription prices or a conversion table for Antigravity AI credits. Google has announced higher prices from 1 December 2026.

Scroll to compare all rate columns.

Gemini 4 Argon · USD per 1 million tokens · paid API tier
Price period	Mode	Input	Cached input	Output + thinking
Through 30 Nov 2026Introductory rate	Standard	$2.00	$0.10	$10.00
Through 30 Nov 2026Introductory rate	Batch	$1.00	$0.05	$5.00
Through 30 Nov 2026Introductory rate	Flex	$1.00	$0.05	$5.00
Through 30 Nov 2026Introductory rate	Priority	$3.60	$0.18	$18.00
From 1 Dec 2026Announced rate	Standard	$4.00	$0.20	$20.00
From 1 Dec 2026Announced rate	Batch	$2.00	$0.10	$10.00
From 1 Dec 2026Announced rate	Flex	$2.00	$0.10	$10.00
From 1 Dec 2026Announced rate	Priority	$7.20	$0.36	$36.00

Argon has no free API tier. Cache-hit tokens use the cached-input rate instead of the regular input rate. Tool charges can apply separately; check Search and Maps grounding prices before estimating a complete workflow.

For existing API integrations, remove explicit cache creation and review output limits. Argon ignores temperature, top-p, and top-k settings. The migration guide covers the request changes.

Choose for the task

Google's thinking-level guidance

Start with a representative task and clear acceptance checks. These are editorial starting points, not a benchmark ranking; compare total time, token use, and the quality of the finished work.

Task guidance; model and thinking-level choices depend on your client and plan
Your task	Start with	Why / when to switch
Small edits and quick iterations	Gemini 3.8 Flash	Try the broadly available fast option first, then compare quality on your own code.
Complex features and agent workflows	Gemini 4 Argon · medium	Google recommends medium as the API starting level for most tasks, including complex code.
Difficult debugging, math, or science	Gemini 4 Argon · high	Evaluate deeper thinking when medium falls short; allow for more tokens and slower first output.
Compare a second model provider	Claude Sonnet 5.5 or Opus 5.5	Run the same acceptance checks. Access needs paid Pro, Ultra, or an enabled Enterprise project.

Argon's API offers low, medium, and high thinking, with medium as the default. The client model picker controls the choices in Antigravity. API limits and defaults do not imply that every client exposes the same settings.

Sources: Antigravity model picker, Argon thinking guidance. For Anthropic's direct API, see the Claude guide.

Plans and usage limits

Antigravity plan documentation

Antigravity uses workload-based quotas. Google does not publish a guaranteed number of prompts per plan: a difficult agent task can consume more allowance than a short edit.

Scroll to compare plans. Subscription prices vary by region, tax, storage tier, and offers; verify your checkout price.

Individual Antigravity access through Google AI plans; Free and Plus share the access tier shown here
Plan detail	Free / Google AI Plus	Google AI Pro · paid	Google AI Ultra
Subscription price	Free: $0; Plus: local tier price	$19.99/month (US reference)	Check current tier and local price
Baseline refresh	Weekly refresh	Every 5 hours, until the weekly cap	Every 5 hours
Weekly allowance	Baseline weekly quota	Higher weekly quota	Highest weekly quota
Gemini 4 Argon	Not included	Included; trial excluded	Included
Beyond baseline quota	No purchased-credit fallback	Purchased or one-time promotional AI credits	Purchased or one-time promotional AI credits

Google's public plan page lists Pro at $19.99 per month in USD. Plus and Ultra have tier-dependent offers; use current Google One pricing for your account. Google One's Gemini usage multipliers are not fixed Antigravity prompt allowances.

Baseline and overagesPaid Pro and Ultra can spend purchased or one-time promotional AI credits after baseline quota runs out. The overage setting lets you choose Never or Always. Charges follow Gemini Enterprise consumption pricing.

Check usage in the clientView model quota in Settings. In the CLI, use /usage for quotas and /credits for the credit panel; credit fallback is controlled in /settings.

Enterprise accessConnect through an eligible Gemini Enterprise Standard or Plus seat, or a Cloud project with consumption billing. Partner Claude models need admin opt-in and token billing; background Gemini coordination can also incur charges.

Sources: baseline quotas and overages, Pro price, credit purchase eligibility, CLI controls, Enterprise setup and billing.

Sources & updates

Last reviewed on 18 October 2026. Access, quotas, retirement dates, and prices can change; check the official sources before choosing a plan or estimating spend.

This independent reference covers Antigravity and Gemini 4 Argon's direct API. For models billed through GitHub, use the GitHub Copilot guide.

Back to all resources