Google releases Gemini 4 Argon
The new model is available in the Gemini app, while Google cites additional safety measures for cyber use.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
Google made Gemini 4 Argon available in the Gemini app on Tuesday. The new version is accessible there through paid AI subscriptions, while Google is also offering the model for business and development applications.
According to Google, Gemini 4 Argon is intended for complex tasks, including software development and extended assignments. In the Gemini app, users can deploy the model for tasks in Google services and in more than forty connected applications.
Google also mentions a direct link with the agentic programming environment Antigravity in the macOS app. In task mode, Gemini can, according to the company, carry out assignments while the user is away from the computer. These are Google's product claims; an independent assessment of performance is not part of the announcement.
Google DeepMind published a model card at the same time. It contains internal safety tests, manual tests by specialist teams and assessments of risks relating to biology, cyberattacks and influence, among other areas.
For biological information and cybersecurity, Google says it cannot rule out Argon reaching a critical capability threshold. The company therefore says it has drawn up additional safety safeguards and measures to bring the remaining risk below that threshold.
The model card also states that the results come partly from automated tests and are not directly comparable with earlier model cards. Google further says that the model is still being evaluated in certain areas of risk.
The launch thus shows two sides of the development. Argon is gaining broader access and more possibilities for carrying out tasks independently, but Google does not present the safety assessment as proof that all risks have been resolved. For users and businesses, the quality of execution, control over actions and handling of data remain important open questions.
One story, several perspectives
What is established
- Google has made Gemini 4 Argon available to certain users and for certain applications.
- Google cites capabilities for software development, task execution and links with other apps.
- Google cannot rule out certain biological and cybersecurity risks.
- The safety assessment is largely based on Google's own tests and model documentation.
Left
Arguments Powerful AI systems should only become more widely available if independent regulators can monitor how they handle privacy, labour, disinformation and cyber risks. Companies should not determine safety rules entirely by themselves.
Values Public oversight, protection of workers and citizens, privacy and precaution.
Consequences A slower pace of introduction may come at the cost of innovation, but reduces the risk that social harm only becomes visible after large-scale use.
Centre
Arguments A phased introduction with model cards, red teaming, logging and clear liability creates room for innovation without abandoning oversight. External evaluations should supplement company claims.
Values Proportionality, transparency, innovation and institutional oversight.
Consequences A combination of access and oversight can build trust, but requires sustained investment in audits and incident response.
Right
Arguments European and American companies should retain room to compete quickly and take advantage of new productivity gains. Excessive regulation could result in safe applications being developed elsewhere.
Values Technological progress, entrepreneurship, competitiveness and individual freedom of choice.
Consequences Faster introduction can deliver economic benefits, but places more responsibility on companies and users to manage risks themselves.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The launch, features and safety assessments are taken directly from Google's publications and attributed cautiously. The text makes no independent performance claim.
- confirmed Gemini 4 Argon was made available in the Gemini app on 13 October. — Google dates the announcement to 13 October 2026. source
- confirmed Google AI Pro and Ultra subscribers gain access to Argon in the app. — This is stated in Google's product announcement. source
- confirmed Gemini is linked to more than forty applications. — Google cites this number in the announcement. source
- confirmed Google cannot rule out Argon reaching a critical cybersecurity threshold. — The model card contains this wording and describes additional safety measures. source
- confirmed The model card uses partly automated tests. — Google DeepMind describes the testing method and warns of limited comparability. source
Editor's note
The availability and safety claims come from Google itself. Independent performance benchmarks or an external audit have not been published in the sources consulted.Sources
- Gemini 4 Argon comes to the Gemini app — Google
- Gemini 4 Argon Model Card — Google DeepMind
- MIPCOM Cannes 2026 — MIPCOM
More on this in Dutch media
- NOS — „google gemini”
- Het Parool — „google gemini”
- NRC — „google gemini”