On 13 October 2026, Google made its frontier model Gemini 4 Argon generally available. Developers and companies can use it through the paid Gemini API in AI Studio and the Gemini platform. The model ID is gemini-4-argon. Google AI Ultra subscribers can also try it in Google Antigravity. In the Gemini app and AI Mode in Search, access is for Google AI Ultra and Pro subscribers.
This article stays within Google’s public posts: who can use Argon, what it costs, and which jobs Google says it fits. It also lists checks before you put it on your own work.
What it costs to try
Google Cloud lists an introductory price of $2 per million input tokens and $10 per million output tokens through 30 November 2026. Cached input is 95 percent off. From 1 December, standard pricing is $4 input and $20 output per million tokens. Google’s developer post says you start on the paid tier of the Gemini API. A yen list price is not in the announcement, so budget in dollars and remember that output costs more than input.
What Google says it is for
Google describes Argon as a model for hard problems in coding, finance, legal work, cybersecurity, and science. It says Argon ranks first on the Vals Index, which covers finance, coding, legal, and tax work. Harvey’s head of applied research said results improved over Gemini 3.1 Pro on corporate M&A, data privacy and cybersecurity, and venture work, including long documents. ServiceNow reported about 20 percent higher task success than Gemini 3.8 Flash in early access, with gains in HR and customer service. Those are the companies’ own evaluations.
Examples from inside Google
Google says teams using Argon in Antigravity increased agentic code submissions by more than 35 percent and cut rollbacks by nearly a third. The Chrome team used it to refactor a legacy folder of about 6 million lines inside Chromium’s roughly 40 million-line codebase, finishing months of work in weeks. On the open-source Agent Development Kit, an overnight run cut in-memory session prep by 40 percent in long sessions and sped cold imports by 21 percent. Google also reports up to 2.3x faster molecular-structure prediction in xTB, and 1.8x faster conformer search. These figures are Google’s, not an independent rerun.
Security is framed as defense
Google Cloud says Argon blocks offensive tasks such as pentesting and exploitation, while defensive use goes through tools such as CodeMender. It ties for first on CWE-Bench v1 at 68 percent for fixing vulnerabilities. On Gray Swan’s indirect prompt-injection benchmark, Google says attack success across 15 adaptive attempts fell to 0.7 percent. An unmitigated build is limited to vetted defenders in the Fairwind Program. The public API is not a license to run attacks.
What to check before you adopt it
- Price both the introductory rate and the December rate
- Decide where contracts, ledgers, and customer data are sent
- Have a person check citations and calculations against the source
- Watch hidden instructions when the model reads email or the web
- Merge generated code only after tests and review
A first-place benchmark is not a first-place result on your own documents. Keep a human stop on irreversible steps, then compare Argon with the model you already run on the same inputs.
Summary
Gemini 4 Argon opened on 13 October 2026 on the paid Gemini API, and for eligible Gemini app, AI Mode, and Antigravity users. Introductory pricing runs through 30 November 2026 at $2 / $10 per million tokens, then $4 / $20. Primary source: Google’s blogs.
Makoto Tejima builds generative AI, RAG, and AI agents into existing business systems and web services. Contact us if you want help choosing a model, designing the human check, and setting permissions.