Google on October 14 opens Gemini 4 Argon frontier model globally with introductory API pricing through November 30

Google CEO Sundar Pichai and DeepMind head Demis Hassabis jointly announced on the early morning of October 14 Beijing time that Gemini 4 Argon, the first frontier model in the Gemini 4 family, is now generally available globally to developers and enterprises. The model targets software engineering, cybersecurity, financial and legal work, quantum computing, and life sciences as complex multi-step tasks, with the per-task output limit raised from 64,000 tokens in the prior generation to 1,000,000 tokens so the model can reason, generate, self-check, and correct itself within a single turn without handing it back to the user. Gemini 4 Argon is available through the Gemini API, AI Studio, the Gemini platform API on Google Cloud, and Google Antigravity, while Pro and Ultra subscribers can use it directly in the Gemini App and AI Mode in Search, and Antigravity Ultra subscribers get the highest access tier. More than 100,000 Google employees already use the model internally, with engineering team agentic code submissions up 35% and post-rollback rates down nearly a third over the past four weeks; the Chromium team used the model to shorten a multi-month code refactor into weeks. On safety, external bio red-team task unlock rates are below 1%, indirect prompt-injection attack success is down to 0.7%, and coding shortcut attempts are down to 2%. Google is also expanding the Fairwind cyber defender program to hundreds of partners and giving 10,000 scientists six months of free highest-tier Antigravity access.