Skip to main content

Gemini 4 Argon: Write a Million Tokens in One Go

Google DeepMind's New Model Can Write a Million Tokens Without Stopping

Google DeepMind quietly dropped Gemini 4 Argon on September 30, and the headline number is hard to ignore: a single output can now reach 1 million tokens. That's not the context window—it's how much the model can generate in one continuous run. For tasks like migrating a massive codebase, writing a long technical report, or automating a multi-step process, this means no more breaking things into chunks. Most cutting-edge models top out around 128,000 tokens, so Argon's leap is a big deal.

Image

Built for the Long Haul, Not Small Talk

Argon isn't for everyone right away. Google is rolling it out in phases: first to trusted cybersecurity professionals, then to developers, enterprises, and subscribers. The focus is on "complex, long-term workflows"—software engineering, law, finance, and cybersecurity defense. And the benchmarks back it up. DeepSWE v1.1, which tests long-term software engineering tasks, hit a record 77.9%. Vals Index, which weighs finance, programming, law, and taxation, ranks Argon first. Zapier's AutomationBench gave it 51.3%, also first. Long video understanding? LVBench came in at 91.7%, the highest yet. On the visual side, it can handle professional charts, pull details from long videos, and take actions across documents.

Security Is the Main Event

Cybersecurity is the core story here. Argon is trained to autonomously find, verify, and fix critical vulnerabilities. For trusted defenders and Google's internal teams, there's even a version with no network barrier. Security vendor Wiz used it through the Scan for Good project to spot a serious flaw in a global hospital's medical software—something previous cutting-edge models had missed. On CWE-bench v1, which tests vulnerability repair, Argon ranked first with 68%.

Pricing and Defenses

Pricing comes in two tiers: during the intro period, it's $2 per million input tokens and $10 per million output tokens. After that, it jumps to $4 and $20. Before making it widely available, 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 Argon to debug code, design algorithms, and handle large-scale code migrations. One team improved quantum algorithms by 40% in minutes with its help.

In short, Argon isn't built for casual chat. It's built to tackle the hardest problems professionals face.

Key Points

  • 1 million token output: A single run can generate up to 1 million tokens, far beyond the typical 128K limit.
  • Targeted release: First to cybersecurity pros, then developers, enterprises, and subscribers.
  • Top benchmarks: Leads in long-term software engineering (77.9%), Vals Index, AutomationBench (51.3%), and LVBench (91.7%).
  • Security focus: Autonomously finds and fixes vulnerabilities; already caught a hospital software flaw.
  • Pricing: $2/$10 per million input/output tokens initially, rising to $4/$20 later.