Skip to main content

Gemini 4 Argon: 1M Token Output, Built for the Hard Stuff

Google DeepMind's Gemini 4 Argon: A Model That Doesn't Quit

What if an AI could write a million tokens in one go—no breaks, no stitching? That's exactly what Google DeepMind's new Gemini 4 Argon promises. Launched on September 30, it's not your average chatbot. It's a workhorse for the gnarliest tasks out there.

The Output Limit That Changes Everything

Most cutting-edge models today cap output at around 128,000 tokens. Argon? It blows past that with a staggering 1 million token output limit. That's not about how much you can feed it—it's about how much it can produce in a single, uninterrupted chain. Imagine migrating an entire codebase, writing a lengthy technical report, or automating a multi-step process without the model losing its train of thought. That's the kind of continuity Argon brings to the table.

Built for the Long Haul

Google isn't just dropping this model on everyone at once. It's a phased release: first for trusted cybersecurity pros, then developers, enterprises, and subscribers. The focus is on "complex, long-term workflows" in software engineering, law, finance, and cybersecurity defense. And the benchmarks back it up:

  • DeepSWE v1.1: 77.9% on long-term software engineering tasks—a record.
  • Vals Index: Ranks first, weighted across finance, programming, law, and taxation.
  • Zapier's AutomationBench: 51.3%, topping the chart.
  • LVBench (long video understanding): 91.7%, the highest score yet.

On the visual side, Argon can handle professional charts, extract details from long videos, and perform cross-document actions. It's not just about text—it's about understanding complex, multi-modal information.

Security: Defense First

Cybersecurity is at the heart of this release. Argon is trained to autonomously discover, verify, and fix critical vulnerabilities. For trusted defenders and Google's internal teams, there's even a "no network barrier" version. In one case, security vendor Wiz used Argon through the Scan for Good project to spot a serious vulnerability in a global hospital's medical software—something previous cutting-edge models had missed. On the CWE-bench v1 vulnerability repair test, it ranked first with 68%.

Pricing and Access

During the introductory period, Argon costs $2 per million input tokens and $10 per million output tokens. After that, it'll be $4 and $20, respectively. Before wide release, Google is beefing up defenses: preventing abuse, blocking prompt injection, monitoring misaligned reasoning chains, and strengthening sandboxes.

Thousands of Google employees are already using Argon to debug code, design algorithms, and perform large-scale code migrations. In one case, quantum algorithms were improved by 40% in minutes with its help. Clearly, Argon isn't for casual conversation—it's for professionals tackling the hardest challenges.

Key Points

  • 1 million token output limit: Enables uninterrupted generation of massive content, from codebases to long reports.
  • Top benchmark scores: Leads in software engineering, automation, and long video understanding.
  • Phased rollout: Starts with cybersecurity experts, then expands to developers and enterprises.
  • Security focus: Designed to autonomously find and fix vulnerabilities, with a defense-oriented approach.
  • Two-tier pricing: Introductory rates of $2/$10 per million tokens, later $4/$20.

Image

Argon is a glimpse into the future of AI: not a jack-of-all-trades, but a master of the long, complex, and critical.