Google releases Gemini 4 Argon for complex real-world tasks

Gemini 4 Argon is live globally today: developers and enterprises can call it via AI Studio and the Gemini platform API, Google AI Ultra subscribers get it in Antigravity, and Ultra and Pro subscribers get it in the Gemini app and AI Mode in Search. Pricing is $2 per million input tokens and $10 per million output tokens, with a 95% cached-input discount, as introductory pricing. Google says it ranks first on the Vals Index, covering work scenarios in finance, coding, law, and tax. In internal engineering practice, the Chrome team used Argon to separate a ~6-million-line legacy feature catalog from core web-processing logic, completing an architectural refactor across a 40-million-line Chromium codebase in weeks instead of months and involving hundreds of files; Argon in Antigravity raised agent code commits by over 35% and cut code rollbacks by nearly a third. The open-source ADK Python codebase, after nightly autonomous optimization, reduced long-session memory preparation time by 40% and sped up cold-start imports by 21%; xTB molecular structure prediction sped up by up to 2.3x, and Gemma 4 E2B on A100 with vLLM achieved 2.85x end-to-end inference speedup with identical tokens. On safety and research, Google is extending the model to hundreds of Fairwind Program partners, and Wiz used it to find and fix vulnerabilities that could have leaked government sensitive data and healthcare infrastructure master keys. Isomorphic Labs used it to compress weeks of combined patent and PDF extraction into hours; in quantum computing, Argon turned an internal version of AlphaQubit into a single optimized hardware kernel, nearly doubling inference throughput. For teams running open-source models, these upstream changes have been submitted to the official xTB and google/adk-python repositories.