Follow the latest coverage, related explainers and connected technology stories.
Google presents the Gemini 4 Argon model as an advanced model that outperforms GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 in 14 of 19 internal benchmarks. However, independent evaluations reveal weaknesses in some programming tasks, while the model's public availability remains tied to safety testing.
Webrazzi reviewed a selection of the most notable artificial intelligence developments of the week, from incidents involving agent behavior and the voluntary safety agreement in the United States to new models and products that perform real tasks on behalf of users. Growing investments and infrastructure commitments reveal that the shift from chatbots to agents is simultaneously raising the ceiling for capabilities, costs, and risks.
Google reviewed a set of artificial intelligence updates it announced during September 2026, led by the Gemini 4 Argon model with a context window of one million tokens and the Fairwind program for testing its security capabilities. The updates included audio models, new integrations with applications, tools for Windows, and research projects for mapping the genome and tracking methane from space.