Google DeepMind Announces Gemini 4 Argon AI Model
Google DeepMind announced its new frontier AI model, Gemini 4 Argon, which handles complex tasks in coding, knowledge work, and cybersecurity. The model is currently rolling out to trusted cyber defenders through the Fairwind Program.
In short: Google DeepMind announced its new frontier AI model, Gemini 4 Argon, which handles complex tasks in coding, knowledge work, and cybersecurity. The model is currently rolling out to trusted cyber defenders through the Fairwind Program.
Artificial intelligence is taking another big step forward as Google DeepMind unveils its latest system designed to tackle long and complicated workflows in one go.
What happened, in plain words
Google DeepMind announced Gemini 4 Argon, a new AI model built for deep reasoning across software engineering, finance, legal, and cybersecurity tasks. Argon is currently being rolled out to a set of trusted cyber defenders through the Fairwind Program and is undergoing a phased release process. The model handles complex workflows and features an output token limit of 1M tokens. Internally at Google, Argon has helped optimize quantum algorithms, freed up memory across data centers, and assisted with large-scale code rewrites.
Key points
- Phased Release and Pricing Argon is rolling out to trusted cyber defenders through the Fairwind Program. It launches at a price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off.
- Internal Engineering Success Thousands of Googlers are already using Argon for coding and research. It optimized quantum computing subroutines 40% faster than the published baseline and freed up over 300 TiB of memory in data centers.
- Expanded Output Limit Google expanded the model's output token limit to an industry-leading 1M tokens, up from 64K tokens, giving the model more room to think deeply and solve tough problems in a single try.
- Cybersecurity Defense Argon can autonomously find, validate, and patch software vulnerabilities. Wiz used Argon to uncover a critical vulnerability in healthcare software used by hospitals worldwide.
- Safety and Safeguards Google is strengthening safeguards against misuse and prompt injection attacks, using internal and external red teams to test the model's robustness before a broader release.
Terms explained
- Tokens — Pieces of words or text that AI models read and generate. Example: When you type a sentence, the computer breaks the words down into smaller chunks called tokens to understand them.
- Quantum Algorithm — A set of instructions written for advanced quantum computers to solve specialized problems. Example: Researchers use these specialized rules to make super-powerful computers run difficult calculations much faster.
- Red Teams — Groups of people who test a system by trying to break it or find its weaknesses. Example: A security company might hire a group of ethical hackers to act as a red team and try to break into their computer network.
- Prompt Injection — A trick where hidden malicious instructions are used to hijack an AI's behavior. Example: An attacker might slip secret commands into a website document to force a chat assistant to break its safety rules.
Why it matters
As AI models become better at long tasks like coding and data analysis, tools like Gemini 4 Argon may help engineers build software faster and assist professionals with complex financial or legal research.
What we still don't know
Argon is still in an early testing phase with restricted access, and Google is actively gathering feedback and iterating on guardrails before making the model available to all developers, enterprises, and consumers.
Based on reporting from Google DeepMind. This is an independent explainer, written in our own words with AI assistance; Google DeepMind has not reviewed or endorsed it. Read the original for the full details.