Google has released Gemini 4 Argon globally for developers and enterprises through its API and for Google AI Ultra subscribers. The model is also reaching Google AI Ultra and Pro subscribers in the Gemini app, giving paying users access to the company’s latest frontier system.
API pricing starts at $2 per million input tokens and $10 per million output tokens, while cached inputs receive a 95% discount. Google positions Argon for multi-step work in coding, finance, law, cybersecurity and science, but many of the performance examples come from Google’s own teams and early-access customers.
The company says Argon helped engineers separate browser-handling code from a roughly six-million-line Chromium folder and complete a refactor across hundreds of files. Google also reports that the model found changes now used in its open-source Agent Development Kit, cutting long-session preparation time by 40% and cold-import time by 21%.
Those examples show the kind of sustained engineering work Google wants the model to handle, rather than guaranteeing the same results on every codebase. Developers can now test the claims directly and weigh the model’s higher reasoning depth against token cost and the need to review high-impact output.