Google Launches Gemini 4 Argon With Focus on Reasoning and Cybersecurity
publishers 19articles 19first reported 30 Sep, 20:11 UTCdeveloping since 1 Oct · 2 editions
Google has launched Gemini 4 Argon, a new artificial intelligence model designed for sustained reasoning across finance, software engineering, coding, creative writing and cybersecurity defense. The Wall Street Journal reported that Google is rolling out the model gradually amid safety concerns, while The Guardian reported that Google is restricting access over safety concerns.
Google said Argon can analyze charts, long-form videos and collections of documents, and can autonomously identify, validate and patch critical software vulnerabilities. The company also said it added protections against prompt injection and measures intended to prevent the model from acting without a user’s direction.
Artificial Analysis ranked Argon alongside OpenAI’s GPT-6 Astra on its composite Intelligence Index, according to Engadget. The benchmarking company also reported that Argon had a 15% hallucination rate and tied with Grok 4.7 and GPT-6 Astra for first place on the CWE-bench cybersecurity leaderboard.
Google is offering introductory pricing of $2 per million input tokens and $10 per million output tokens. Argon has an output limit of 1 million tokens, Engadget reported.
The company said it has already used Argon for quantum-computing research, codebase migrations and data-center memory optimization, where the model helped free 300 tebibytes of memory. Google also said an early demonstration uncovered a critical vulnerability in healthcare software used by hospitals worldwide. The release follows Gemini 3.5 and comes as Bloomberg reported employee skepticism about the new model and The New York Times reported its release with guardrails amid an A.I. safety debate.
HOW THIS STORY WAS MADE
Written from 8 articles, headlines and summaries only; 8 independent newsrooms once syndicated copies count as one; this version written 8 h after the record first saw the story; the editor revised it.
- Publishers
- 19
- Source articles
- 19
- Given to the writer
- 8, headlines and summaries only
- Independent newsrooms
- 8
- This version written
- 8 h after the record first saw the story
- Second model (editor)
- revised
- Publication gate
- passed
- Human review
- none
Written by a language model from the sources above, then checked by a second model that may only cut, attribute or correct. How it works →