Google announces Gemini 4 Argon without wide release
The new model is already being used in internal workflows and is initially being tested with a limited group.
Google has announced Gemini 4 Argon as a new model for complex tasks such as programming, research and cybersecurity. The model is not yet widely available to developers, businesses or consumers.
Google DeepMind writes that Gemini 4 Argon is already being used in internal workflows. Thousands of employees are said to be testing the model for, among other things, specialised programming tasks, in-depth research and writing texts.
Google is also enabling a limited group of cybersecurity experts to work with the model. The announcement is therefore primarily an introduction to a technology and a testing programme, not to a product that everyone can use immediately.
Google presents Argon as a model capable of handling longer and more complex workflows. In its own description, the company mentions software development, knowledge work, legal and financial applications, and cybersecurity. Those applications are the maker's claims; independent comparisons are still limited.
The company also refers to tests against indirect prompt injection, in which a model is misled by instructions hidden in documents or web pages. Google says Argon performs strongly in its own Gray Swan test. Without a full test setup, independent replication and comparison with other models, such a company benchmark says little about everyday use.
The fact that the model is not yet being released generally temporarily reduces the direct consequences for users in the Netherlands. The announcement is nevertheless relevant for businesses because they must assess whether new models are reliable enough for sensitive business information, software code and decision-making.
The introduction fits into a broader race for models that not only generate text but also carry out lengthy tasks. The key questions therefore concern not only speed or performance, but also control, data use, liability and which tasks people will continue to monitor.
One story, several perspectives
What is established
- Google has announced a new AI model.
- The model is initially being tested on a limited basis.
- The available performance claims come largely from Google itself.
Left
Arguments Advanced AI should primarily be assessed by its consequences for workers, public services and the concentration of power. Without robust rules, companies can privatise the benefits and pass the risks on to staff and society.
Values Equality, public control, and protection of labour and privacy.
Consequences Stricter requirements may delay launches, but reduce risks in sensitive applications.
Centre
Arguments New models can be useful, but only with verifiable safety standards, human oversight and transparency about data and performance. The limited testing phase provides room to establish such conditions first.
Values Innovation with oversight, reliability and institutional responsibility.
Consequences A phased rollout takes time, but can increase trust and safe deployment.
Right
Arguments Companies should retain room to innovate quickly and compete globally. Liability and misuse must be addressed, but rules must not become so burdensome that European organisations fall behind.
Values Entrepreneurship, technological progress and competitiveness.
Consequences Faster introduction can boost productivity, but increases the need for clear accountability afterwards.
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 text distinguishes between Google's own claims and independently established facts. The limited availability and announcement have been confirmed by Google and Axios.
- confirmed Google announced Gemini 4 Argon. — Official announcement by Google. source
- confirmed The model is being used internally and by a limited group of cybersecurity experts. — Google describes internal workflows and a select programme for cyber defence. source
- confirmed The model is not yet widely available to consumers and developers. — Axios describes the limited introduction; Google positions it as an announcement and testing phase. source
- confirmed Google says that Argon performs strongly in a Gray Swan test for prompt injection. — This is explicitly a claim from Google's own announcement. source
Editor's note
Google has announced the model and describes internal and limited testing. Independent performance and safety assessments are still limited; company claims have been presented as such.Sources
- Gemini 4 Argon: our next era of frontier intelligence — Google DeepMind
- Presentamos Gemini 4 — Google
- Google unveils long-awaited Gemini 4 — Axios
More on this in Dutch media
- Tweakers — „google gemini”
- AD — „google gemini”
- RTL Nieuws — „google gemini”