Google has launched Gemini 4 Argon, its first major frontier model since Gemini 3 in late 2025. The company says Argon improves reasoning and complex workflows, and performs better on tasks that can expose familiar chatbot weaknesses such as hedging and repetition.
The release puts Google back into a crowded contest among companies building advanced AI models. Its claims, early uses inside Google, price and limited initial availability offer a first look at how the company plans to compete.
Google pitches Argon as a model for demanding work
Google describes Argon as delivering “frontier reasoning and improved performance across complex workflows.” It says the model reduces hedging, repetition and chatbot cliches, positioning the release as an effort to make responses more useful as well as capable.
The company says Argon outperforms GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 across benchmarks in coding, cybersecurity, science, finance and law. Those are Google’s benchmark claims; the source article does not provide scores or describe the test methods, so the results here cannot be compared in detail.
Google also says it is already using Argon internally. Examples include finding and patching cybersecurity vulnerabilities, reducing by half the computing power needed for YouTube’s comments section, and debugging the drone physics simulator for Alphabet’s drone delivery company Wing.
Together, the examples suggest Google sees applications beyond chat: software development, system efficiency and technical simulation. They are company-reported uses, and the announcement does not provide further details about how the work was measured.
Access starts with higher-tier subscribers
Argon is rolling out to Google’s Pro and Ultra subscribers in the Gemini app and AI mode in Search. Ultra subscribers can also use it on Antigravity, Google’s agentic development platform.
Google has not made clear when, or whether, the model will reach Plus subscribers or people using free accounts. For now, access depends on a higher-tier subscription, and the launch does not establish a broader release timetable.
The company previously limited access to a select group of “trusted cyber defenders” to test the model for potential risks. That staged approach echoes the cautious rollout described in the source, while the current release gives some paying users access through Google’s products.
Low token prices come with a comparison caveat
Argon’s introductory price is $2 per million input tokens and $10 per million output tokens. The source article says that is around a fifth the cost of OpenAI’s GPT-6 Astra and Anthropic’s Fable 5 models.
Those introductory prices are set to double at the end November, when the opening period finishes. Even after that increase, the article says Argon will remain significantly cheaper than frontier rivals on a per-token basis.
A lower token rate does not automatically mean a task will cost less overall. Models can consume different numbers of tokens to complete the same work, so a fair comparison depends on both the rate and the amount used.
A launch that changes the competitive picture
Google’s previous major frontier release was Gemini 3 in late 2025. In the months after that launch, the source says models from Anthropic and OpenAI impressed users while Google fell behind, and internal shakeups raised doubts about its ability or desire to compete at the frontier.
Argon’s arrival is a sign that Google remains in the race, though the company’s claims will need to be considered alongside actual performance, availability and cost in use. Its release came a day after Meta announced Muse Wonder Full, a reminder that the group of companies vying to build frontier AI is broader than two rivals.
For users, the immediate questions are practical: whether Argon handles complex tasks well, how much it consumes to do them, and when access might expand. Google has supplied its performance claims and starting prices, but the first rollout is limited to higher-tier subscribers.