Google has opened Gemini 4 Argon to the public, placing a new frontier model in the hands of developers, businesses, and higher-tier subscribers. The launch gives Google a major answer to OpenAI and Anthropic, and its pricing makes the move impossible to ignore.
Argon arrives with claims of strong performance across complex work, from coding and cybersecurity to science, finance, and law. Google also launched it at a cost that sits around a fifth of OpenAI’s GPT-6 Astra and Anthropic’s Fable 5 models, at least during the introductory period.
Gemini 4 Argon Moves Beyond Its Preview
Google released Gemini 4 Argon to the general public after an initial two-week preview limited to cybersecurity partners and internal teams. The model is now available through the paid Gemini API tier, enterprise customers, Google AI Ultra, and selected Gemini app and Search modes.
Google is also rolling Argon into the Gemini app and AI mode in Search for higher-tier Pro and Ultra subscribers. Ultra subscribers can access it on the Antigravity platform. Google has not made clear when, or if, Argon will reach Plus subscribers or free users.
The model’s first major public release follows Gemini 3 in late 2025, giving Google a new chance to show that it can keep pace with the frontier. Internal shakeups had raised doubts about Google’s ability to compete at that level, especially as Anthropic and OpenAI released impressive models in the months after Gemini 3.
Argon’s arrival sends a direct message: Google remains in the AI race. The company launched the model just one day after Meta announced its frontier model, Muse Wonder Full, turning this release into part of a crowded stretch for top AI companies.
A Powerful Model With an Aggressive Price
Google launched Argon with an introductory price of $2 per million input tokens and $10 per million output tokens. Those prices will double after November 30, but Argon will still remain significantly cheaper than its rivals on a per-token basis.
- Input tokens cost $2 per million during the introductory period.
- Output tokens cost $10 per million during the introductory period.
- Both prices will double after November 30.
- The introductory cost is around a fifth of GPT-6 Astra and Fable 5 pricing.
That price structure could make Argon attractive for teams that process large amounts of information or run complex workflows through an API. Lower token costs give developers more room to test demanding applications, while enterprise customers can examine the model without facing the same price level attached to rival systems.
The value proposition becomes stronger after the introductory offer ends. Even with prices doubled, Google says Argon will still undercut its rivals on a per-token basis, giving the model a cost advantage that could shape adoption as companies compare performance and budgets.
Google Targets Complex Workflows and Real-World Systems
Google says Argon delivers “frontier reasoning and improved performance across complex workflows.” The company also says the model reduces hedging, repetition, and cliches, aiming for responses that stay focused when tasks demand more than a simple answer.
Google claims Argon outperforms rival models across benchmarks covering coding, cybersecurity, science, finance, and law. Those areas point toward a model built for technical and professional work, not only everyday questions or basic content generation.
The company has already used Argon internally for several demanding jobs. It helped find and patch cybersecurity vulnerabilities, cut the computing power needed for YouTube comments in half, and debug a drone physics simulator.
Each example highlights a different kind of challenge. Security work tests the model’s ability to identify weaknesses and help repair them, while the YouTube comments task shows a direct computing-power gain. Debugging a drone physics simulator pushes Argon into technical systems where errors can affect how a virtual machine behaves.
These internal uses also give Google a strong way to frame the release. Argon is not arriving as a model that exists only in benchmark tables; Google presents it as a system already helping with security, large-scale platform workloads, and simulation debugging.
Google’s Frontier Challenge Is Now Public
Argon enters a contest that includes GPT-6 Astra, Claude Fable 5.1, Claude Opus 5.5, and Meta’s Muse Wonder Full. Google says its new model can compete with the industry’s best, raising the stakes for every company trying to lead advanced AI.
The timing matters because Google’s rivals gained ground after Gemini 3 arrived in late 2025. Anthropic and OpenAI released impressive models, while Google faced internal changes that created doubts about its frontier ambitions. Argon now gives the company a public product that can answer those doubts through access, pricing, and claimed benchmark performance.
The next test will come as developers and enterprise customers use Argon beyond Google’s own systems. Its rollout reaches paid API users, enterprise customers, Google AI Ultra, Pro and Ultra subscribers, and selected Search experiences, while its low introductory price puts the model in front of teams deciding which AI platform deserves their workloads.
Google has put a new contender on the field, and Argon’s combination of frontier claims and lower token costs could make the competition much more intense. The model’s public release turns Google’s AI comeback question into a live contest over performance, access, and price.
Based on
