Google Launches Gemini 4 Argon, Its Most Advanced AI Model — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Alphabet unveiled Gemini 4 Argon on Wednesday, claiming it sets a new record in real-world software engineering and ties for first in cybersecurity benchmarks.
- Gemini 4 Argon ties OpenAI's GPT-6 Astra and Grok 4.7 on cybersecurity benchmarks and outperforms GPT-6 Astra and recent Anthropic models on the Vals Index measuring finance, legal, and other professional tasks.
- Google is already running Argon internally at its data centers, where the model has freed up hundreds of terabytes of memory without additional hardware purchases.
- Sundar Pichai signed a voluntary AI safety agreement with President Donald Trump and other major tech executives at the White House a day before the launch, following a luncheon on rising AI safety concerns.
- Google will roll out Argon in phases, starting with trusted cybersecurity partners while completing pre-release safety evaluations with the U.S. government across four areas including misuse and prompt injection.
Why it matters: Google reasserted itself at the frontier by tying OpenAI and Grok on cyber while beating them and Anthropic on professional benchmarks, and it's already validating Argon on its own infrastructure — the model freed hundreds of terabytes of memory at Google data centers without new hardware. The gated, defender-first rollout, paired with Pichai's White House safety pledge, shows Google is using national-security framing to justify why customers can't yet touch its most capable model.
Ask SkimNews
