Google has released Gemini 4 Argon, its first major frontier model since Gemini 3 arrived in late 2025, putting the company back into direct competition with the latest systems from OpenAI and Anthropic. The model began rolling out on October 13 to higher-tier subscribers in the Gemini app and AI Mode in Search, while Google’s Antigravity development platform is also receiving access, The Verge reported.

Google is presenting Argon as a reasoning model for complex workflows rather than a narrow update to its consumer chatbot. The company says the system improves performance across coding, cybersecurity, science, finance and law while reducing hedging, repetition and formulaic language. Those comparisons are based on Google’s claims and benchmark results cited at launch; they do not by themselves establish how the model will perform across real-world tasks.

The model is entering products at different access levels. The Verge reported that Google Pro and Ultra subscribers can use Argon in the Gemini app and Search’s AI Mode. Ultra subscribers also receive access through Antigravity, Google’s agentic development environment. Google had not clarified whether Argon would reach Plus subscribers or free users.

One AI engine branches into chat, search and software development environments.
Google is distributing Argon across the Gemini app, AI Mode in Search and Antigravity.

Pricing is one of Google’s clearest competitive moves. Argon launched with an introductory API rate of $2 per million input tokens and $10 per million output tokens. The Verge calculated that this was about one-fifth the listed price of OpenAI’s GPT-6 Astra and Anthropic’s Fable 5 models. Google plans to double Argon’s rates after the introductory period ends at the close of November.

Even after that increase, the listed per-token rates would remain below those rivals, according to The Verge. But token price is not the same as the cost of completing a task. Models may consume different amounts of context, reasoning tokens or generated output to reach a comparable result, so a lower unit price does not guarantee that every workload will be cheaper.

Google says it is already using Argon internally. Examples cited by the company include finding and patching cybersecurity vulnerabilities, reducing by half the computing power required for YouTube’s comments section and debugging the physics simulator used by Wing, Alphabet’s drone-delivery company. The Verge’s report did not include independent verification of those internal results or enough technical detail to reproduce them.

Different token streams pass through AI engines toward the same completed task.
Lower token prices do not automatically mean a lower total cost for every workload.

The public release followed a smaller rollout to selected cybersecurity professionals described as trusted cyber defenders. That staged approach gave specialists early access to examine potential security risks before the broader launch, matching a pattern increasingly used for powerful models with cyber capabilities. Developers can now reach Argon through Google’s model interfaces as the consumer and search rollouts expand.

Argon arrives after months in which newer systems from Anthropic and OpenAI had raised questions about Google’s position at the frontier. The company also faced internal changes that fueled uncertainty over how aggressively it would compete. Launching a model across Search, the Gemini app and an agentic coding platform shows Google using its distribution advantages rather than treating Argon as a stand-alone API product.

The result is a three-part wager: that Argon’s reasoning quality can match the leaders, that lower prices can attract developers, and that Google’s existing products can put the model in front of users quickly. The first two claims will become clearer only after independent testing and sustained production use. What is already clear is that Google intends to compete on both model capability and the cost of deploying it.