TL;DR
On Wednesday 30 September Google announced Argon, a new top-tier model that heads its Gemini 4 family. A limited set of cybersecurity partners get it first, and there is no public release date. The long-promised Gemini 3.5 Pro has been dropped altogether.
What Google is claiming
According to a company spokesperson, Argon is bigger than Google’s earlier “Pro” models and is “our most performant model yet built for complex workloads”. On the coding and cyber tests it treats as most important, Google ranks Argon alongside Astra from OpenAI and Anthropic’s Opus.
The self-reported numbers are mixed. Argon beat both rivals on several industry tests in Google’s press release, Reuters reports, yet it still lagged on others, including two of the four coding benchmarks Google chose to publish. Vendor-selected benchmarks deserve caution at the best of times; a release that concedes ground on half its own coding scores is unusually candid.
How Google got here
In July we reported that Google had shipped three cheaper Gemini models while its flagship slipped. That flagship, Gemini 3.5 Pro, was originally promised for June by chief executive Sundar Pichai, and Google no longer plans to release it. In the months between, DeepMind was reorganised: founder Demis Hassabis stepped aside as chief executive, and several leaders of the Gemini effort left Google.
Reuters says the delays let Anthropic and OpenAI pull further ahead. Pichai rejected that framing in July, but Google’s pitch has drifted from frontier capability towards price, presenting Gemini as the cheaper option.
The testing question
Google says it is taking part in the Trump administration’s voluntary scheme for pre-release model access. Reuters does not say whether any UK body is included. That matters here: last week Washington pressed AI labs to let US agencies test new models before Britain’s AI Security Institute, and Argon’s pre-release route runs through the American scheme.
Looking forward
For UK businesses on Google Cloud, nothing changes until Argon has a public date and pricing. The more telling signals will be whether AISI publishes any evaluation of the model, and whether Google’s cost-led positioning survives the arrival of a larger flagship.