> ## Content Index > Fetch the complete content index at: https://www.tao.media/llms.txt > Use this file to discover other available public pages before exploring further. # Google Launches Gemini 4 Argon Frontier Model for Coding and Cyber Defense - URL: https://www.tao.media/google-launches-gemini-4-argon-frontier-model-for-coding-and-cyber-defense/ - Published: 2026-10-01T21:33:21.000Z - Updated: 2026-10-01T22:23:42.000Z - Description: The model is rolling out first to trusted cyber defenders before a wider release to paid API customers and Google AI Ultra subscribers. - Author: Tristan Hillerich - Tags: Google, AI, News [Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/?ref=tao.media) has introduced Gemini 4 Argon, a new frontier AI model built for long-horizon coding, enterprise knowledge work, and cybersecurity defense. The model is being released in phases, beginning with a limited group of trusted cyber defenders through Google DeepMind’s [Fairwind Program](https://deepmind.google/fairwind-program/?ref=tao.media). Google said it is also participating in the U.S. government’s voluntary pre-release model access process while it gathers feedback from early testers and strengthens safeguards ahead of broader availability. Argon is expected to become available to developers, enterprises, and consumers later, starting with paid API customers and Google AI Ultra subscribers. Google describes Argon as its next step in frontier reasoning, emphasizing models that can sustain longer tasks, analyze larger working contexts, and operate across more complex professional workflows. Introductory pricing is set at $2 per million input tokens and $10 per million output tokens. Cached input tokens will be priced at a 95% discount. After the introductory period, Google said pricing will move to $4 per million input tokens and $20 per million output tokens. ## Gemini 4 Argon Expands Long-Horizon Reasoning The headline technical change in Gemini 4 Argon is its output token limit. Google said the model supports up to 1 million output tokens, up from a previous 64,000-token ceiling. Many advanced model workflows now depend less on short answers and more on extended reasoning, code generation, debugging, document analysis, planning, and repeated tool use. A higher output limit gives the model more room to work through multi-step tasks in a single trajectory, rather than compressing reasoning or handing work back to the user after a short response window. Argon is already being used internally at Google across specialized coding tasks, deeper research, and writing workflows. The company cited examples from quantum algorithmic optimization, data center memory efficiency, and large-scale code migration work. In one internal quantum computing example, Google said Argon improved a published baseline by 40% while optimizing spacetime resources, measured as qubits multiplied by gates, for subroutines that bottleneck important applications. In another case, a team of Argon agents analyzed fleet-wide profiling telemetry and identified memory optimizations across Google data centers, with more than 300 TiB of memory expected to be freed after rollout and an estimated 500 TiB to 1 PiB in total savings. Argon agents are being used in large C/C++-to-Rust migration work, including codebases ranging from core libraries such as re2 and libgav1 to larger systems such as the Fuchsia Zircon kernel. The company said those migrations are subject to automated and manual auditing, emulation testing, and review before production use. ## Coding, Finance, Legal, and Automation Benchmarks Google’s launch materials frame Gemini 4 Argon as a model for work that crosses coding, reasoning, multimodal understanding, and domain-specific professional analysis. For software engineering, Google reported a 77.9% score on DeepSWE v1.1, a benchmark focused on real-world, long-horizon software engineering tasks. Beyond coding, the company said Argon leads on the [Vals Index](https://www.vals.ai/benchmarks/vals%5Findex?ref=tao.media), which measures economic-impact work across finance, coding, legal, and tax categories weighted by their contribution to U.S. GDP. Google also cited leading performance on Vals Finance Agent v2 for multi-step financial research and Harvey’s Legal Agent Benchmark for legal research and drafting. On Zapier’s AutomationBench, which measures end-to-end execution across business workflows, Google said Argon ranked first with a score of 51.3%. ![Gemini 4 Argon benchmarks](https://storage.ghost.io/c/78/0b/780ba906-b1a7-4bf0-873c-bdd5c32e5331/content/images/2026/10/gemini4-argon-benchmarks-1.png) The model is also being positioned for multimodal knowledge work. Google said Argon can analyze charts, process long videos, and act across document sets. On LVBench, a long-video understanding benchmark, the company reported a 91.7% score. Google frames Argon as a model for extended professional workflows rather than a general chatbot update, with users asking it to reason across code, documents, visual material, and domain-specific tasks over longer periods. ## Google Opens Full Cyber Capabilities to Trusted Defenders First Cybersecurity is the most controlled part of the rollout. Google said Gemini 4 Argon was trained for defensive cybersecurity tasks, including finding, validating, and patching critical software vulnerabilities. For trusted defenders and internal Google teams, the company said Argon will be released without cyber guardrails so approved users can access its full frontier cyber-defense capabilities. That decision reflects a difficult balance for frontier AI labs. A model that can autonomously identify vulnerabilities may help defenders patch serious exposures faster, but similar capabilities can also be misused if released broadly without controls. Google’s first-stage release through the Fairwind Program keeps the most sensitive cyber functionality limited to vetted users while the company continues testing safeguards. Google cited early work with Wiz’s [Scan for Good](https://www.wiz.io/scan-for-good?ref=tao.media) initiative, which provides free scanning and remediation support for critical public infrastructure. According to Google, Argon identified a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide, including a severe risk previous frontier models had missed. On [CWE-bench v1](https://cwe-bench.com/?ref=tao.media), a benchmark for remediating security vulnerabilities, Google said Argon tied for first with a top score of 68%. The company also said Argon showed improvements over 3.8 Flash Cyber on internal vulnerability discovery tests and on Wiz’s internal black-box penetration testing benchmark.