- Published on
Google Begins Gemini 4 Argon Frontier Model Rollout
- Authors
- Name
- DeepHorus
Google has begun a broad rollout of Gemini 4 Argon, its new frontier model, making it available to developers, enterprises and consumers worldwide and moving beyond the early-access and demonstration phase. The release marks the first general availability of a model Google has positioned as its most capable system for professional and scientific work. Argon can be accessed through the paid Gemini API, AI Studio and the Gemini platform, as well as through Google Antigravity for Google AI Ultra subscribers. Google AI Ultra and Pro subscribers can also reach the model via the Gemini app and AI Mode in Search.
Google set introductory pricing at $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95%. Through the Gemini API, the model supports a unified input-output context window of up to 1 million tokens, designed for long-running scientific, coding and agentic tasks. Access is also being extended to specialized users: researchers can apply for six months of the highest Google Antigravity usage tier at no charge, subject to eligibility and capacity. Cybersecurity professionals can seek access under separate security arrangements through Google's Fairwind Program.
In coding, Google said Argon ranks first on the Vals Index, a benchmark covering professional work across finance, coding, legal and tax. Internally, Google engineers using Argon through Antigravity increased agent-generated code submissions by more than 35% while cutting code rollbacks by nearly one-third. The company also reported that Argon helped restructure portions of Chromium's 40 million-line codebase, compressing work that would have taken months into weeks. On Google's open-source Agent Development Kit, Argon-generated changes reduced in-memory session preparation time by 40% in long sessions and improved cold-import performance by 21%.
Google highlighted Argon's contributions to research, including improvements to quantum error-correction software, work on nuclear-fusion mathematics, and assistance with rare-disease genome analysis and enzyme design. To support such work, the company is offering 10,000 scientists access to the highest usage tier of Antigravity for six months, subject to eligibility and capacity.
Robotics featured prominently in Google's disclosures, with Argon scoring 87% on the company's Embodied Reasoning benchmark, which tests reasoning about physical and spatial problems in 3D environments. In an internal test, Argon watched video of a person playing cornhole, mapped the scene, calculated required body movements and generated simulation code that was transferred to Apptronik's Apollo humanoid. The robot made seven consecutive shots in a real-world trial, according to Google. Google described the approach as a "Video-to-Sim-to-Real" pipeline intended to cut robot-training simulation creation from weeks to hours.
Google also introduced added safeguards around agent behavior, prompt injection and potentially harmful requests. The company has framed the 1 million-token unified window as central to Argon's ability to work across extended coding, research and agentic tasks, suggesting continued emphasis on long-horizon autonomy as the model reaches a wider audience.
