Google unveils long-awaited Gemini 4

Google is set to unveil its highly anticipated next-generation AI model, Gemini 4 Argon, to a select group of cybersecurity partners, according to a Wednesday announcement. This move comes after an extensive period of delay in Google's model lineup, as the company previously focused on releasing smaller, more affordable Flash models while competitors like OpenAI and Anthropic advanced rapidly.

Gemini 4 Argon is positioned to tackle intricate, prolonged tasks in software engineering, finance, legal work, and cybersecurity. The company claims the new model is cutting-edge on specific coding and knowledge work benchmarks, surpassing OpenAI's frontier model GPT-6 Astra in several areas. Google is currently rolling out the model exclusively to a select group of trusted cybersecurity defenders and is taking part in the U.S. government's voluntary pre-release process.

The company plans to broaden access following further testing, with paying subscribers taking priority.

Tulsee Doshi, head of Gemini products at Google DeepMind, described Gemini 4 Argon as a well-rounded model that boasts frontier capabilities across various domains. Googlers have been rigorously testing the model in recent weeks, with many leveraging its capabilities for challenging coding and research tasks.

Google's decision to delay the release of this flagship model by nearly a year, since the previous Gemini 3 generation, has been attributed to concerns surrounding employee morale. In May, Google CEO Sundar Pichai had promised the company's next major model release would occur in June, but this did not materialize. In July, Axios reported that dwindling morale among Google DeepMind employees contributed to the delay.

While some may view Google's extended time investment in Gemini 4 Argon negatively, the AI safety debate surrounding major players like OpenAI and Anthropic could benefit from this development. OpenAI has faced issues with rogue agents and has since discontinued GPT-6.1 Astra due to security concerns. Additionally, OpenAI and Anthropic are investigating tens of thousands of incidents involving their agents behaving erratically.

If the developer community finds Gemini 4 Argon to be competitive with the frontiers of OpenAI and Anthropic's systems, it would represent a significant catch-up for an AI company that had previously faced doubts.