Google Gemini 4 Argon Launched
Google has announced Gemini 4 Argon, a new frontier artificial intelligence model designed to handle complex and long-running tasks across several professional fields. The company says the model is focused on real-world software engineering, enterprise knowledge work, legal and financial workflows, and cybersecurity defence.
The announcement marks Google's latest move in the competition among major artificial intelligence companies. Gemini 4 Argon is being positioned as a model capable of sustaining deeper reasoning across multi-step workflows rather than focusing only on short question-and-answer interactions.
According to Google, Argon has been developed to work on demanding tasks that require extended reasoning, coding and multimodal understanding. The company has also highlighted its ability to assist with cybersecurity activities, including finding, validating and patching software vulnerabilities.
One Million Output Tokens
One of the key technical features announced for Gemini 4 Argon is its ability to generate up to one million output tokens. Google says this represents a substantial increase over the previous 64,000-token output limit.
The larger output capacity is designed for tasks that require lengthy responses, extended coding workflows and multi-stage reasoning. However, a larger token limit does not necessarily mean every response will use the full capacity or that longer responses will automatically produce better results.
The feature could be particularly relevant for software development and enterprise workflows where an AI system may need to process or generate large amounts of information during a single task.
Focus on Software Engineering
Google has positioned Gemini 4 Argon as a model for advanced software engineering. Its capabilities include reasoning across code, working through complex programming tasks and supporting longer development workflows.
Google's published benchmark table reports strong results for Argon on several coding and agentic software engineering evaluations. For example, Google reports a 77.9 percent score for Argon on DeepSWE v1.1 and 91.9 percent on Vibe Code Bench. At the same time, the company's own comparison table shows competing models ahead on some other coding evaluations, including Terminal-Bench 4.0 and FrontierSWE v2.
These figures are based on Google's published evaluations and should be considered in the context of the specific testing methods and conditions used.
Applications in Finance and Legal Work
Gemini 4 Argon is also designed for enterprise knowledge work. Google has specifically highlighted applications involving finance and legal work.
The model's reasoning capabilities are intended to help with complex information analysis, research and other professional workflows. Google has published benchmark results covering financial and legal agent tasks as part of its comparison with other frontier AI systems.
For businesses, such capabilities could support workflows that require analysis of large volumes of information and multiple steps before reaching an output. The actual usefulness of the model for professional applications will depend on factors such as accuracy, reliability, security and integration with existing systems.
Cybersecurity Capabilities
Cybersecurity is one of the major areas highlighted in Google's announcement.
Google says Gemini 4 Argon can autonomously find, validate and patch critical software vulnerabilities. The company also said that the model has been used to identify a security vulnerability in healthcare software used by hospitals worldwide.
Because advanced cybersecurity capabilities can also create potential misuse risks, Google has chosen a controlled rollout. The model is initially being provided to a group of trusted cybersecurity defenders through the Fairwind programme.
Google says it is continuing to work on safeguards before making the model broadly available.
Gemini 4 Argon Availability
Gemini 4 Argon is not initially available to everyone. Google has started the rollout with selected trusted cyber defenders through the Fairwind programme.
The company plans to expand access subsequently to paid API customers and Google AI Ultra subscribers, followed by broader availability for developers, enterprises and consumers. The exact timing of each stage can change as Google continues testing the model and its safety systems.
Gemini 4 Argon Price
Google has announced introductory API pricing of 2 dollars per million input tokens and 10 dollars per million output tokens. Cached input tokens receive a 95 percent discount from the standard introductory input price.
The announced introductory pricing is intended for API usage and should not be confused with the price of a consumer subscription. Google has indicated that pricing can change after the introductory period.
For Indian users and businesses, the final cost in rupees will also depend on the applicable exchange rate and taxes.
Comparison With Other AI Models
Google has compared Gemini 4 Argon with several competing frontier models, including OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5.
Google's published benchmark table shows Argon leading on some evaluations while competing models lead on others. For example, Google reports that Argon scores 68.9 percent on the Vals Index Knowledge Work evaluation, compared with 63.1 percent for GPT-6 Astra and 67.0 percent for Claude Opus 5.5. On Terminal-Bench 4.0, however, Google reports 66.4 percent for Claude Opus 5.5 compared with 57.4 percent for Argon.
This means benchmark performance varies depending on the task and evaluation. Results published by a model developer should also be distinguished from independent testing.
What Happened to Gemini 3.5 Pro
The launch of Gemini 4 Argon also changes the expected path for Google's frontier model lineup. Reports indicate that Google had previously discussed Gemini 3.5 Pro but ultimately moved to the Gemini 4 Argon release instead.
The new model therefore represents Google's next major frontier release rather than a conventional update to the previously expected Gemini 3.5 Pro.
What This Means for Users
For developers, Gemini 4 Argon could become relevant for complex programming and long-running agentic workflows once wider access begins. Businesses may also be interested in its applications across legal, financial and enterprise knowledge work.
Cybersecurity organisations are another important potential user group, although Google's initial controlled rollout reflects the additional safety considerations associated with advanced vulnerability discovery and software security capabilities.
Consumers will need to wait for broader availability before they can assess how the model performs in everyday applications.
Google's Next Step
Google is currently focusing on testing Gemini 4 Argon with trusted users and improving its safety systems before a broader release. The company says feedback from early users will help refine the model and its safeguards.
As access expands, independent evaluations will provide additional information about how Gemini 4 Argon performs outside Google's own testing environment. Until then, the benchmark results announced by Google should be understood as company-reported performance measurements rather than a universal ranking of AI models.
Gemini 4 Argon is therefore an important new addition to Google's AI model portfolio, with its main focus on complex reasoning, software engineering, enterprise work and cybersecurity. Its one-million-token output capability, controlled rollout and introductory API pricing are among the major features to watch as the model becomes more widely available.
