[Digital Today reporter Chi-gyu Hwang] Google has unveiled a new AI model, 'Gemini 4 Argon', that improves performance for specialised work such as coding, cybersecurity, finance and law.

CNBC reported on Sept. 30 that Google plans to roll out Argon in stages, starting with cybersecurity partners.

Google said Argon has significantly improved cybersecurity capabilities compared with the 3.8 Flash Cyber model it unveiled earlier this month and has stronger vulnerability detection performance. Tulsee Doshi (툴시 도시) described Argon as a very balanced model that is strong at long, multi-step tasks.

The company said Argon tied with OpenAI GPT-6 Astra and Grok 4.7 on a cybersecurity evaluation benchmark.

It set a new record on an actual software engineering benchmark. It also led on another benchmark that measures overall performance across specialised tasks such as finance and law.

Google said it will conduct safety evaluations with the U.S. government before releasing Argon to the public. It plans to proceed with a public release after expanding safeguards in 4 key areas, including misuse and prompt injection.

Doshi, Google's product lead for the Gemini model, said the initial rollout approach helps build trust while also helping provide a model strong in cyber defence to defenders as quickly as possible.

Argon is already being used in Google's internal operations. Google said it secured hundreds of terabytes of memory without additional hardware by using the Argon model to optimise data centre memory. Quantum computing researchers are also using Argon.