Google launches the company's most powerful Gemini 4 Argon
"Gemini 4 Argon is the highest-performing model we've ever built for complex work, comparable to cutting-edge products on the market like OpenAI's Astra or Anthropic's Opus on key cybersecurity and encryption standards," a Google spokesperson told Reuters after Gemini 4 Argon launched on September 30.
The company claims the new model delivers "outstanding performance" in programming and cyber defense, as well as knowledge work such as legal and finance. This AI achieved the highest score with CWE-bench - a test of AI's ability to find and patch security vulnerabilities, equivalent to OpenAI's GPT-6 Astra, and "set a record score in the performance test in real, long-term technical tasks", but did not give details.
According to Google, company employees currently use Gemini 4 Argon for tasks such as debugging and large-scale source code migration. In it, the new model helps engineers free up more than 300 tebibytes (TiB) of memory across the data center "without the need for new hardware".
Google has not disclosed the release date for Gemini 4 Argon to general users, but said it is "actively cooperating" with the US government, allowing relevant agencies to assess cybersecurity and other risks before deployment. In addition, to control the safe operating model, Google deploys its own system to help monitor Argon's thought chain and block it when necessary.
According to Business Insider, Gemini 4 Argon is the company's latest effort to regain its position in the race to build artificial intelligence, especially with security and programming. Previously, despite launching many versions of Gemini, Google's AI was still rated worse than products from OpenAI and Anthropic.
Google CEO Sundar Pichai said in June that he would introduce Gemini 3.5 Pro with "great power". However, the model was delayed several times due to poor performance and was later unavailable on the market. According to TechCrunch, Gemini 4 Argon is expected to bring Google back to the leading group.
Google's announcement of the new model comes as many competitors launch powerful products. On September 1, Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1, focusing on programming, research and knowledge tasks. OpenAI then introduced GPT-6 Astra on September 3, targeting programming, research, and multi-step tasks. By the end of the month, Anthropic announced Claude Opus 5.5 (September 22) and Sonnet 5.5 (September 28), while OpenAI released GPT-6 Sol, GPT-6 Luna on September 22 and GPT-6.1 Sol on September 29, expanding its ability to handle professional work and programming.