Skip to main content

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

Google DeepMind's Gemini 4 Argon: A Million Tokens, Zero Interruptions

What if an AI could write a novel in one sitting—or refactor an entire codebase without losing its train of thought? Google DeepMind's latest model, Gemini 4 Argon, launched on September 30, does exactly that. Its standout feature? A staggering 1 million token output limit—not the context window, but how much it can generate in a single, uninterrupted response. That's roughly eight times the output capacity of most cutting-edge models today.

Why Output Size Matters

Until now, even the best models capped out around 128,000 tokens per output. That forced developers to break tasks into chunks, losing continuity. Argon changes the game: deep code migrations, multi-step automation, and long technical reports can now flow from start to finish without segmentation. As one engineer put it, "It's like going from writing on Post-it notes to having a whole whiteboard."

Image

Built for Professionals, Not Chatting

Google is rolling Argon out in phases—first to trusted cybersecurity experts, then developers, enterprises, and subscribers. The target? Complex, long-term workflows in software engineering, law, finance, and cyber defense. The benchmarks back it up:

  • DeepSWE v1.1: 77.9% on long-horizon software tasks—a new record.
  • Vals Index (finance, coding, law, tax): ranked #1.
  • Zapier's AutomationBench: 51.3%, also #1.
  • LVBench for long video understanding: 91.7%, the highest yet.

Beyond text, Argon handles professional charts, extracts details from lengthy videos, and performs cross-document actions. It's less about conversation and more about getting hard things done.

Security That Defends, Not Just Detects

Cybersecurity is Argon's core mission. It's trained to autonomously find, verify, and fix critical vulnerabilities. For trusted defenders, Google even offers a "no network barrier" version. In one case, security firm Wiz used Argon through the Scan for Good project to uncover a severe flaw in a global hospital's medical software—something previous models missed entirely. On CWE-bench v1, it ranked first for vulnerability repair with a 68% success rate.

Pricing and Defenses

During the introductory period, Argon costs $2 per million input tokens and $10 per million output tokens. Later, prices will double to $4 and $20. Before wide release, Google is reinforcing four advanced defenses: preventing abuse, blocking prompt injection, monitoring misaligned reasoning chains, and strengthening sandboxes.

Thousands of Google employees already use Argon daily—debugging code, designing algorithms, and migrating massive codebases. In one case, it improved quantum algorithms by 40% in minutes. So no, Argon isn't here to chat about the weather. It's here to tackle the toughest challenges professionals face.

Key Points

  • 1 million token output: Generate entire codebases or reports in one go.
  • Top benchmarks: #1 in long software engineering, finance/law, automation, and video understanding.
  • Phased rollout: Starting with cybersecurity pros, then developers and enterprises.
  • Security focus: Autonomous vulnerability discovery, plus a no-network version for defenders.
  • Pricing: $2/$10 per million tokens initially, doubling later.