SAN FRANCISCO — Google has unveiled Argon, the flagship model for its new Gemini 4 generation, after months of delays that have pushed the company to accelerate efforts to keep pace with OpenAI and Anthropic.
The new model is larger than Google’s previous advanced Pro models and is designed for complex workloads, with the company highlighting its performance in software engineering and cybersecurity.
Google said Argon is being made available initially to selected cybersecurity partners rather than released broadly to the public. The company has not set a timetable for wider availability.
A Google spokesperson described Argon as the company’s most capable model yet and said it was comparable with leading models from OpenAI and Anthropic on selected coding and cybersecurity benchmarks. Google said Argon performed better than those competitors on several benchmarks disclosed by the company, while falling behind on two of four coding tests included in its comparison.
The restricted rollout reflects concerns about the potential misuse of increasingly capable AI systems. Google said Argon is designed to refuse requests that could assist cyberattacks or the development of chemical, biological or nuclear weapons.
The company is also participating in a voluntary US government programme that allows authorities early access to advanced AI models before wider release.
Google’s decision to limit the initial distribution comes after months of delays surrounding its next-generation models. Chief Executive Sundar Pichai had previously indicated that Gemini 3.5 Pro would be released in June, but Google has now abandoned plans to launch that model.
The delay has coincided with a period of organisational change at Google DeepMind, Google’s artificial-intelligence research unit. Several senior figures associated with the Gemini programme have left the company, while the organisation has undergone a broader restructuring.
The setbacks have come as OpenAI and Anthropic have continued releasing increasingly capable models, intensifying competition at the top end of the AI market.
Google has sought to challenge that pressure by emphasising not only model performance but also the cost and efficiency of its AI systems. Pichai has rejected suggestions that Google is falling behind in artificial intelligence, even as the company has adjusted its strategy following the delays.
Argon’s initial focus on cybersecurity also reflects the growing importance of advanced AI in both defending computer systems and identifying vulnerabilities within them. Google said early testing showed the model could detect and address critical software weaknesses that other advanced AI systems had missed.
In one test cited by the company, Argon identified a vulnerability in software used by hospitals that could have exposed sensitive personal information.
Google said it was monitoring the model’s reasoning and behaviour as part of its safety testing. The company is also examining the possibility that highly capable systems could act beyond the intentions of their users, a concern researchers describe as misalignment.
The safety concerns come as governments and technology companies face growing pressure to establish safeguards around increasingly powerful AI systems. The US administration has created a voluntary process for companies developing frontier models to provide authorities with access before public release.
Google’s announcement came shortly after Pichai and other technology executives met US President Donald Trump at the White House, where major technology companies agreed to a voluntary framework aimed at addressing risks associated with artificial intelligence.
Argon’s limited release means Google’s latest flagship model will not immediately be available to ordinary users. The company plans to use feedback from selected partners and internal testing before determining how and when it can be made more widely accessible.
The launch also effectively resets Google’s flagship AI roadmap. Rather than releasing the delayed Gemini 3.5 Pro, the company is moving directly toward Gemini 4 with Argon as its leading model.
For Google, the challenge now is to demonstrate that the new generation can regain ground in a market where rivals have continued to release advanced systems while the company worked through technical delays and changes inside its AI division.
