Skip to main content

Gemini 4 Argon: 1M Token Output, Built for the Long Haul

Google's New AI Model Writes a Million Tokens in One Go

Imagine asking an AI to write an entire codebase migration plan, and it doesn't stop until it's done—no breaks, no "continue" prompts. That's the promise of Gemini 4 Argon, unveiled by Google DeepMind on September 30. Its output limit has leaped from 64K to a staggering 1 million tokens. To put it plainly: that's not about how much it can read (the context window), but how much it can generate in a single go.

Image

Why Output Size Matters

Most cutting-edge models today cap output at around 128,000 tokens. Argon's jump to a million means it can tackle deep codebase migrations, lengthy technical reports, or multi-step automation processes without awkward breaks. The result? Continuity from start to finish. No more stitching together fragments—just one seamless flow.

Built for the Pros

Argon isn't for casual chit-chat. Google is rolling it out in phases: first to trusted cybersecurity professionals, then to developers, enterprises, and subscribers. It's designed for 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 (weighted for finance, programming, law, and taxation): ranked first.
  • Zapier's AutomationBench: 51.3%, also first.
  • LVBench for long video understanding: 91.7%, the highest.

On the visual side, it handles professional charts, extracts details from long videos, and performs cross-document actions.

Security That Defends, Not Just Detects

Cybersecurity is the headline here. 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. Security vendor Wiz used it through the Scan for Good project to spot a serious flaw in a global hospital's medical software—one that previous cutting-edge models had missed. On CWE-bench v1, it ranked first for vulnerability repair with a 68% score.

Pricing and Defenses

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

Thousands of Google employees are already using it to debug code, design algorithms, and perform large-scale code migrations. In one case, quantum algorithms were improved by 40% in minutes.

In short, Argon isn't about chatting—it's about tackling the hardest challenges for professionals.

Key Points

  • 1 million token output: Generate entire projects in one go, no breaks.
  • Targeted release: First for cybersecurity pros, then developers and enterprises.
  • Top benchmarks: Leads in software engineering, automation, and long video understanding.
  • Security focus: Finds and fixes vulnerabilities autonomously; already caught a hospital software flaw.
  • Tiered pricing: $2/$10 per million tokens initially, rising to $4/$20 later.