Gemini 4 Argon is moving beyond its initial limited release. Google announced the model on September 30; October 13 brings an Antigravity rollout, while Google’s current Gemini documentation lists Argon for personal-account Pro and Ultra subscribers. The plan you need depends on which app you want to use. Original announcement, today’s Antigravity rollout.
Argon is aimed at demanding coding, research and professional workflows. For most readers, the practical starting point is checking access. A subscription that includes Argon in the Gemini app does not necessarily unlock it in Google’s coding tools.
Which plans include Argon?
| Where you want to use it | Included access | What to check |
|---|---|---|
| Gemini Apps, personal account | Google AI Pro and Google AI Ultra | Free and AI Plus do not list Argon. Usage allowance and remaining capacity still apply. |
| Antigravity | Google AI Ultra and eligible Enterprise access | Free, AI Plus and AI Pro do not include Argon in Antigravity’s current model table. |
| Gemini API | Separate developer access and billing | Google’s Antigravity rollout announcement lists API availability; consumer subscription entitlements are a separate product. |
The Gemini Apps information above applies to personal Google Accounts. Work and school accounts follow their organization’s license and administrator settings. Google’s model support page lists Argon for both chats and tasks on Pro and Ultra, with selectable thinking levels. The Antigravity model table sets out its different entitlements.
When will Argon appear in Antigravity?
Google says the rollout starts October 13 in the Antigravity desktop app and CLI for Google AI Ultra subscribers and eligible Gemini Enterprise subscriptions. It advises updating to the latest version.
There is a meaningful product distinction: the documentation lists Argon for Antigravity 2.0, the CLI and IDE extensions, and explicitly excludes the separate Antigravity IDE. If you are looking in the old IDE, an absent Argon option does not establish that the wider rollout has failed.
“Rolling out starting today” also does not promise a precise hour for every account. Google does not give an account-specific activation time in the October 13 announcement. Check the app version, eligible plan and model selector before changing subscriptions.
The benchmark scores Google reports
The following is Google DeepMind’s published comparison, checked October 13. These are results from specified tests and settings, rather than a prediction that one model will perform best on every task. Published performance table.
| Benchmark | Argon | GPT-6 Astra | Claude Fable 5.1 | Claude Opus 5.5 |
|---|---|---|---|---|
| Vals Index | 68.9% | 63.1% | 65.8% | 67.0% |
| AutomationBench | 51.3% | 41.4% | 31.4% | 42.5% |
| Vals Finance Agent v2 | 65.4% | 53.5% | 58.9% | 58.6% |
| Harvey’s Legal Agent Benchmark | 19.6% | 5.4% | 6.7% | 3.8% |
| DeepSWE v1.1 | 77.9% | 74.1% | 67.4% | 74.2% |
| FrontierSWE v2 | 55.0% | 65.5% | 56.3% | 62.3% |
| Vibe Code Bench | 91.9% | 89.6% | 90.3% | 90.3% |
| Terminal-Bench 4.0 | 57.4% | 58.2% | 57.9% | 66.4% |
| PostTrainBench | 45.3% | 44.3% | 40.2% | 49.3% |
| Terminal-Bench Science 0.1 | 57.6% | 68.1% | 52.6% | 63.3% |
| LABBench 2 | 88.8% | 85.4% | 68.6% | 73.1% |
| RiemannBench | 76.0% | 72.0% | 65.6% | 69.6% |
| GraphWalks, up to 128K, BFS F1 | 99.7% | 98.7% | 91.4% | 90.6% |
| GraphWalks, 256K–1M, BFS F1 | 84.2% | 71.8% | 65.0% | 66.8% |
| Agent’s Last Exam, pass rate | 39.5% | 34.2% | — | 38.2% |
| OSWorld 2.0, offline partial score | 69.2% | 72.6% | — | — |
| Chartography | 71.6% | 71.0% | 46.2% | 66.3% |
| LVBench | 91.7% | 87.5% | 79.7% | 83.7% |
| CWE-bench v1 | 68.0% | 68.0% | 58.0% | 67.0% |
“—” means no comparison score is supplied in Google’s table.
The results show a mixed picture. Argon leads this comparison on several knowledge-work and coding measures, but other models score higher on FrontierSWE, Terminal-Bench and some science or computer-use tests. That is a reason to evaluate your own workflow rather than treat a launch headline as a universal ranking.
Google’s evaluation methodology says the Argon results generally use single-attempt evaluation and the API’s highest thinking settings. Some tests have different harnesses, budgets or special settings. Non-Gemini scores largely come from provider reports or public leaderboards; this is not one independent test lab running every model identically.
The million-token claim and API pricing
Google’s announcement highlights a one-million-token output limit. That should not be casually rewritten as a promise that every Gemini or Antigravity subscription exposes a million-token chat context.
The announcement lists introductory API prices of $2 per million input tokens and $10 per million output tokens, with cached input at a 95% discount from the input price. It lists $4 input and $20 output after the introductory period. These are the announced API rates, separate from consumer subscription pricing; check current developer pricing before budgeting a workload. Google’s pricing announcement.
What I would check before paying more
Start with the product you actually use. If your goal is Argon chat or tasks, check your Gemini account’s model selector. If your goal is coding in Antigravity, check that product’s eligible plan and supported app. I would test a few representative jobs before upgrading solely for the model name: a coding agent’s usefulness depends on how it handles your repository, review process and permissions as well as its benchmark scores.
This report uses Google’s announcements, support documentation and published evaluations. We have not independently reproduced the benchmark results or verified rollout completion for every account.