Google opens Gemini 4 Argon worldwide to developers, enterprises and subscribers
Google first announced Gemini 4 Argon on September 30 and limited it to trusted cyber defenders in its Fairwind Program. On October 13, 2026 it began a global rollout: the model is now available on the Gemini API's paid tier, in enterprise products, in the Gemini app and in AI Mode in Search. An introductory API price applies until November 30, 2026.
From a closed cyber-defender preview to general availability in 13 days
Argon was unveiled on September 30, 2026, but at that point only a group of vetted security defenders inside Google's Fairwind Program could use it. The October 13 post, signed by Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google, changed that. In his words, "Argon is rolling out today globally", covering developers, enterprises and consumers.
Google describes Argon as the opening model of the Gemini 4 family and says it intends to bring out the rest of the family. Kavukcuoglu writes that "Argon excels at deep work in complex, real-world tasks", pointing to coding, finance, law, cybersecurity and science. Two companion posts went up the same day, one for the Gemini app and one for developers.
Who gets access where, and which results Google highlights
Access is split by audience. Developers work through AI Studio and Android Studio on the Gemini API's paid tier. Enterprises can try the model in Gemini for business, through the API and in Google Antigravity. Google AI Pro and Ultra subscribers find Argon in the Gemini app and in AI Mode in Search, and Ultra subscribers can also use it in Antigravity and AI Studio. On the API side Google announced an introductory price, with a discount on cached input tokens; once that period ends on November 30, 2026, a higher rate applies to the public-access Argon model. The Gemini API offers a single 1 million token limit that counts input and output together, and the developer documentation describes a new continuation protocol for it.
Every performance figure here comes from Google itself. The company says Argon ranks first on the Vals Index, which covers finance, coding, legal and tax work. It scored 87% on ERQA, a test of physical and spatial reasoning. On Gray Swan's indirect prompt injection benchmark it held attack success to 0.7% across 15 adaptive attempts. The September 30 post had already reported 77.9% on DeepSWE v1.1, a long-horizon software engineering test. Google also says its own teams using Argon inside Antigravity raised agentic code submissions by more than 35% and cut code rollbacks by nearly a third.
A cheaper window for agent pilots, and a different model behind consumer answers
Argon is the first Gemini 4 model, and Google is positioning it for long, multi-step agent work. Because the introductory price runs until November 30, businesses building agents or automations can test it at a lower cost during this window. Google is also reaching out to science and security: it is giving 10,000 scientists the top usage tier in Antigravity free for six months, subject to eligibility, capacity and selection, and it is beginning to extend the Fairwind Program to hundreds of new partners; defenders who are admitted gain staged access to Argon with its cyber guardrails lifted.
The consumer change matters too. In the Gemini app, Google says Argon answers more directly, hedges and repeats itself less, and reports progress more often on longer jobs. The app now links to more than 40 services, among them Instacart, OpenTable, Adobe and Shopify, and a tasks toggle lets Gemini keep working while the user is away. Antigravity's agentic coding interface has also arrived in the Gemini app for macOS. Since Argon is landing in AI Mode at the same moment, for Pro and Ultra subscribers the model behind the AI Mode answers they see when researching a brand is changing as well. We covered an earlier Google model release in our Gemini 3.7 Flash article.
The sources say nothing about Türkiye, so this section is commentary
None of the four Google posts contains anything about Türkiye. They describe the rollout as global but do not name individual countries or supported languages. What follows is therefore UNALSOFT commentary, not sourced fact.
Here is what remains open: when Argon will show up in the Gemini app and AI Mode for users in Türkiye, how well it handles Turkish, whether free users will get it, and whether the API will ever have a free tier for it. The sources describe developer access only through the paid tier and app access only for Pro and Ultra subscribers. Our view: a business in Türkiye planning an agent or automation pilot should mark November 30 on its calendar as the end of the introductory period, and should not put the model in front of live customers before testing it on its own Turkish content. Since all benchmark results are Google's own claims, a small trial on your own business data will tell you more than these numbers.
The UNALSOFT view
We read this less as a model race and more as a timing question for businesses. When a model built for long, multi-step work becomes widely available, it is a reasonable moment to try handing a repetitive process to an agent: drafting proposals, summarising documents or routing customer requests. Start small, though: one process, one measurable goal, and steps that need human approval clearly marked. To keep the workflow from breaking the next time the model changes, we recommend building agents around testable tasks rather than a specific model. That is why our Agentic AI work picks the process first and the model second.
Which of your processes could an AI agent take over?
A short call is enough to pick one repetitive workflow together and design a small, measurable agent pilot around it.