Gemini 4 Argon must reverse Google’s AI inertia

Google has unveiled Gemini 4 'Argon', a state-of-the-art AI model, but it's currently only accessible to a select group of trusted cyber defenders during rigorous testing. Google claims the model will be released to a wider audience "as soon as possible." This marks a significant moment for Google, as many had written off DeepMind as a "frontier lab," with some analysts stating "for all intents and purposes, we believe DeepMind is no longer a frontier lab."

Gemini 4 'Argon' is a substantial improvement over its predecessor, Gemini 3 Pro, released in November 2025. It's poised to challenge OpenAI and Anthropic, with benchmark results indicating it performs on par with GPT-6 Astra and Claude Opus 5.5. Notably, Gemini 4 shows promise in agentic tasks and boasts cybersecurity alignment, as well as a substantial 1 million token output limit, enabling it to generate large bodies of work in one prompt with reduced truncation.

However, the real test for Gemini 4 'Argon' lies in real-world applications. Sources within Google suggest the model struggles with certain coding tasks, despite excelling in benchmarks. The model's pricing is also a concern; it begins at $2/$10 per million input/output tokens, the same as OpenAI's Sol 6.1, but will eventually double to match the $4/$20 price of Claude Opus 5.5. While still cheaper than Astra or Fable, it must demonstrate its worth to users to regain trust in the AI community.

Additionally, Google has enlisted Henry Cavill to promote Googlebook, a move that raises questions about the practicality of the endeavor. Other highlights include a recent Sideload episode featuring Allison Johnson of The Verge discussing AI agents, fuzzy little creatures, and the existential dread of life in 2026. Sadly, Android Central, a blog where the reporter worked from 2016 to 2021, has been shut down by Future, the company that acquired it in 2020.