Google’s new model is arriving across its developer tools and paid services, with the company pitching it for complex coding, research and business tasks.
Google on Tuesday began rolling out Gemini 4 Argon, a new model aimed at developers and organizations handling complex, multi-step work. The launch reaches beyond Google’s API: Argon is available in AI Studio and Gemini’s platform, while access through Antigravity and other Google products depends on subscription or eligibility.
For developers, the release is particularly relevant to teams already building with Google’s cloud and coding tools. Google says Argon can be accessed through the paid Gemini API tier and AI Studio, and has expanded its availability to Android Studio and Antigravity on macOS, Windows and Linux. The company also introduced an Android command-line plugin for agentic workflows. Google AI Ultra subscribers can use Argon in Antigravity; the company says eligible Gemini Enterprise customers also have access.
Google is positioning Argon as a model for work that requires sustained reasoning across coding, finance, legal tasks, cybersecurity and science. It says the model ranks first on the Vals Index, a benchmark it describes as measuring performance across sectors including finance, coding, legal and tax work. That is Google’s characterization of its own launch, not an independent assessment of how the model will perform in customers’ day-to-day systems.
The company’s strongest practical case is its internal software work. Google reports that, over four weeks, more than 100,000 Alphabet users using Argon in Antigravity increased agentic code submissions by 35% and reduced post-submit rollbacks by 30%. Those figures come from Google; the company did not provide an independent comparison in the launch material.
Google also says Argon helped its Chrome team refactor code and assisted engineers working on the open-source Agent Development Kit, a framework for building agents. The launch describes the changes as helping with large-scale engineering and performance work, but the available claims do not establish how the results compare with other models or conventional developer tools.
The model’s ambitions extend well beyond programming. Google highlights work in scientific computing, cybersecurity and biological research, including claimed contributions to quantum chemistry software and enzyme design. It also describes a robotics demonstration in which Argon interpreted video of a person playing cornhole, generated simulation code and helped transfer the task to an Apptronik Apollo humanoid robot. Such demonstrations suggest possible uses; they are not evidence that the model can reliably perform comparable tasks across real-world conditions.
The higher stakes make safeguards part of the product story. Google says it has built in defenses against harmful requests, prompt injection and agents taking unauthorized shortcuts. It reports a 2% rate of shortcut attempts on coding benchmarks and says its defenses reduced success in an indirect prompt-injection test. These are company-reported evaluations, and organizations considering autonomous workflows will still need to set their own permissions, review processes and security controls.
Google lists an introductory API price of $2 per million input tokens, with cached input tokens discounted by 95%. It is offering 10,000 scientists the highest Antigravity usage tier free for six months, subject to eligibility, capacity and selection. The company says the offer is intended to support research and that applications are open.
For teams building on Google’s platform, Argon’s arrival adds another option to the growing market for AI-assisted coding and agent development. The rollout makes the model available across more of Google’s ecosystem, but its broader value will depend on independent testing, dependable performance and whether the gains hold up outside Google’s own examples.

