Google Unveils Gemini 4 Argon, Its Frontier AI Model With a 1M-Token Output Limit, Starting With Cyber Defenders

Alphabet (GOOGL) unveiled Gemini 4 Argon on September 30, calling it its most advanced artificial intelligence model yet, with major gains in coding, cybersecurity and complex professional work. Google pitches it as its next era of frontier intelligence, built for deep reasoning over long, multi-step workflows, and it arrives after the company cancelled Gemini 3.5 Pro and spent the summer shipping smaller Flash models. The output token limit jumps to 1 million tokens from 64K, letting a single response run to hundreds of thousands of tokens.

On benchmarks, Google reports a 77.9% score on DeepSWE v1.1, ahead of Anthropic's Claude Opus 5.5 at 74.2% and OpenAI's GPT-6 Astra at 74.1%. Argon ties for first on the CWE-bench v1 vulnerability-remediation test at 68%, ranks first on Zapier's AutomationBench at 51.3%, and ties GPT-6 Astra and Grok 4.7 on cybersecurity evaluations while leading the Vals Index of finance, legal and other professional tasks. Inside Google, Argon agents found memory optimizations that free more than 300 TiB across its data centers, with 500 TiB to 1 PiB in total savings expected, and are migrating C and C++ code to Rust, including more than 800,000 lines of the Fuchsia Zircon kernel.

The rollout is deliberately staged. Trusted cyber defenders get the model first, some without cyber guardrails, while Google runs pre-release safety evaluations with the U.S. government and hardens defenses against misuse and prompt injection. The launch lands a day after CEO Sundar Pichai signed a voluntary AI safety agreement at the White House. Google AI Ultra subscribers and paid API customers are next, at an introductory $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 afterward. What to watch: the general-availability date, whether the benchmark lead holds in independent testing, and how the pricing lands against GPT-6 Astra and Claude in enterprise deals.