Google Launches Gemini 4 Argon for Complex Coding Tasks

Google has publicly released Gemini 4 Argon, its latest frontier AI model, following a limited September launch to trusted testers. CEO Sundar Pichai announced global availability for Google AI Pro and Ultra subscribers via the Gemini app and API.

The model is engineered for long-horizon agentic workflows, specifically software engineering, cybersecurity, and complex enterprise knowledge work. Google DeepMind SVP Koray Kavukcuoglu stated Argon excels at deep reasoning and structural precision required for these tasks.

“A business case built on the launch rate is a business case built to be rewritten in seven weeks.”

Google reported internal engineers used Argon to refactor a legacy Chromium folder of roughly six million lines within a 40-million-line codebase, completing the work in weeks instead of months. The company claims teams using the Antigravity coding interface increased agentic code submissions by over 35% and reduced rollbacks by nearly a third.

Introductory pricing is set at $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached input. These rates are scheduled to double to $4 and $20 respectively after November 30, with a shared 1 million-token limit for input and output.

Analysts warn enterprises to budget at the higher post-November rates immediately. Greyhound Research chief analyst Sanchit Vir Gogia noted procurement cycles typically exceed the seven-week discount window, and reasoning tokens billed at output rates make per-token pricing a poor guide for agent costs.

On the independent Artificial Analysis Intelligence Index, Argon ties OpenAI's GPT-6 Astra at a score of 53 with high reasoning enabled. It trails Anthropic's Claude Opus 5.5 and Sonnet 5.5 by approximately five points. The launch follows recent releases from Anthropic, OpenAI, and Meta's Muse Wonder.