Google 面向全球开发者与企业发布 Gemini 4 Argon

Google has launched Gemini 4 Argon, a new AI model for complex coding, scientific research, cybersecurity and enterprise workflows, with a one million-token context window.

The company said Argon ranked first on the Vals Index, which evaluates AI model performance across economic sectors, including finance, coding, legal and tax work. The model is priced at $2 per million input tokens and $10 per million output tokens, with cached input tokens available at a 95% discount.

Google said the model is built to handle complex, multi-step tasks and work across large codebases, documents and datasets. It can also use tools over extended periods to complete tasks with less human intervention.

The company added that Argon helped its engineering teams increase agentic code submissions by more than 35% while reducing code rollbacks by nearly a third. In one project, the model helped the Chrome team refactor code across hundreds of files within weeks, work that the company said would otherwise have taken months.

Google said Argon is being used for legal research, financial analysis, scientific discovery and cybersecurity. The model can analyse documents, connect information across sources and carry out multi-step workflows.

Google said Argon can help security teams identify vulnerabilities and analyse codebases. The company is expanding access to the model through its Fairwind Program, which provides cybersecurity capabilities to trusted defenders.

Cloud security company Wiz used Argon to identify vulnerabilities affecting government infrastructure and healthcare systems. Google said one assessment helped identify a flaw that exposed a master key and could have allowed administrator access to healthcare infrastructure.

In scientific research, Google reported using Argon to support work in quantum computing, nuclear fusion, rare disease research and enzyme design. Researchers at Gamow Labs used the model to identify a chromosomal rearrangement linked to a rare developmental lung disorder that routine software had missed for years.

Google is also using Argon to improve robotics development. In an internal test, the model watched a video of a person playing cornhole, mapped the surroundings and created simulation code. Google said it then tested on Apptronik’s Apollo humanoid robot, which successfully made seven shots in a row.

Google said it had built safeguards to reduce the risk of misuse, including protections against prompt injection, monitoring of AI agents and measures to prevent unauthorised actions.

The company is offering 10,000 scientists six months of free access to the highest usage tier of Google Antigravity, subject to eligibility and availability.

Google said Argon is the first release in the Gemini 4 family, with further models expected to follow.