Google Gemini 4 Argon

Google present Gemini 4 Argon, a new frontier AI model that the company says outperforms or equals its competitors in 13 of 18 published benchmarks. The model is initially available only to trusted cybersecurity experts and select pre-release testers, with broader availability planned for later.

Google doesn’t claim that Gemini 4 Argon wins in every evaluation criterion. However, in the scoreboard included in the company’s materials (which were confidential until released), Argon scores higher — or ties for first place — in more categories than GPT-6 Astra or Claude Opus 5.5.

See more articles from iGuRu.gr when you search for news on Google.

iGuRu.gr preferred source

Of the 18 criteria Google released, Argon clearly leads in 12 and is tied for first place in one. GPT-6 Astra clearly leads in three and is tied with Argon in one. Claude Opus 5.5 clearly leads in two. This makes Argon the leading state-of-the-art model based on the total number of top scores in Google’s benchmark set, although the results show a competition that remains marginal, with OpenAI and Anthropic maintaining advantages in several important technical categories.

Argon appears to have the broadest top-performance profile among the three leading models in Google's comparison, but GPT-6 Astra remains ahead in various software tasks, scientific terminal applications, and computing, while Claude Opus 5.5 excels in terminal agent-type workflows and post-training processes.

For enterprise customers, this means that model selection still depends on workload, even though Argon now offers Google the strongest argument for its dominance in the high-end model category, based on the number of firsts in reviews.

Argon also expands Google’s content production cap. The company says the model supports the production of 1 million tokens — an industry-leading feat — up from the previous 64.000 tokens. This is a notable feature for tasks like agentic software engineering, audits, system migrations, and legal reviews, where a model’s value often depends on how long the chain of tasks it can maintain before handing over control to a human again.

newsbot

Although the press releases will range from very select to rare, I said I'd pass...because sometimes the editors hide.