Gemini 4 Argon: our next era of frontier intelligence

2026-10-01 · Google DeepMind

Gemini 4 Argon: Our Next Era of Frontier Intelligence

Google has announced Gemini 4 Argon, a new frontier AI model designed to sustain deep reasoning across complex, long-horizon workflows. The model delivers top-tier performance in real-world software engineering, enterprise knowledge work (such as legal and finance), and cybersecurity defense.

Phased Release and Pricing

To ensure safety, Google is taking a phased approach to release. Argon is currently rolling out to trusted cyber defenders through the Fairwind Program. Google is actively engaged with the U.S. government’s voluntary pre-release model access process and gathering feedback to iterate on guardrails before making it available to developers, enterprises, and consumers.

The model launches with an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95%.

Transforming Internal Workflows at Google

Thousands of Googlers are already using Argon for specialized coding, deeper research, and writing. It has accelerated several breakthroughs:

  • Quantum Algorithmic Optimization: Argon helped researchers optimize spacetime resources of subroutines, beating a published baseline by 40% in minutes.
  • Memory Efficiency: Argon agents analyzed telemetry across Google’s data centers to autonomously apply memory optimizations, freeing over 300 TiB of memory, with estimated total savings of 500 TiB to 1 PiB.
  • Large-Scale Codebase Migrations: Argon agents are migrating C/C++ codebases to Rust, scaling up to 800,000+ lines for the Fuchsia Zircon kernel. For Google's libgav1 video decoder, agents replaced 32K lines of SIMD code with safe Rust, resulting in a memory-safe decoder that runs 2.7x faster than the previous Rust port.

Tackling Complex Problems with Expanded Limits

To support longer, complex use cases, Gemini 4 Argon significantly expands its output token limit to an industry-leading 1 million tokens, up from 64K. This allows the model to think deeply and generate hundreds of thousands of tokens in a single trajectory to solve tough problems in one go.

Enterprise Workflows and Benchmark Performance

Argon excels across various enterprise domains, setting new state-of-the-art records:

  • DeepSWE v1.1: Scored 77.9% in real-world long-horizon software engineering tasks.
  • Vals Index: The leading model measuring economic impact across finance, coding, legal, and tax work.
  • Domain-Specific Benchmarks: Leads on Vals Finance Agent v2 (financial research) and Harvey’s Legal Agent Benchmark (legal research and drafting).
  • AutomationBench: Ranked #1 with a score of 51.3% for end-to-end business function execution.
  • LVBench: Achieved 91.7% for long video understanding, showcasing strong visual understanding capabilities.

Defensive Cybersecurity

Google trained Gemini 4 Argon to be highly capable at cybersecurity defense, equipping defenders for the new era of cyberattacks. Argon can autonomously find and validate vulnerabilities, though wider access is currently restricted to ensure rigorous testing and safety.

Source