# Gemini 4 Argon — What Changed, Why It Matters **Announced:** September 30, 2026, by Google DeepMind (Koray Kavukcuoglu, SVP Google DeepMind & Chief AI Architect). It is Google's new frontier model, positioned as "our next era of frontier intelligence." ## What changed **1. A massive jump in output length.** Argon expands the output token limit to an **industry-leading 1 million tokens, up from the previous 64K** — giving the model headroom to "think deeply and generate hundreds of thousands of tokens in a single trajectory" for long, multi-step tasks ([Google blog](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon); [VentureBeat](https://venturebeat.com/technology/google-unveils-gemini-4-argon-retaking-benchmark-lead-over-openai-and-anthropic-but-in-limited-release)). **2. Benchmark leadership — but not a clean sweep.** Across the 18 benchmarks Google disclosed, Argon leads outright on 12 and ties for first on one (13 of 18 top scores), vs. GPT-6 Astra (3 outright + 1 tie) and Claude Opus 5.5 (2 outright). Key numbers: - **DeepSWE v1.1** (real-world long-horizon software engineering): **77.9%** (vs. Claude Opus 5.5 74.2%, GPT-6 Astra 74.1%) - **Harvey Legal Agent Benchmark**: **19.6%** (vs. Astra 5.4%, Opus 5.5 3.8%) — one of its clearest wins - **AutomationBench** (Zapier, end-to-end business execution): **51.3%** (vs. Opus 5.5 42.5%, Astra 41.4%) - **Vals Finance Agent v2**: **65.4%**; **LVBench** (long video): **91.7%**; **GraphWalks**: **84.2%** - **CWE-bench v1** (vulnerability remediation): **68%**, tied with GPT-6 Astra - Where it trails: **FrontierSWE v2** (55.0% vs. Astra 65.5%) and **Terminal-Bench Science 0.1** (57.6% vs. Astra 68.1%) — both 10.5-point gaps; Claude Opus 5.5 leads **Terminal-bench 4.0** (66.4% vs. 57.4%) ([VentureBeat](https://venturebeat.com/technology/google-unveils-gemini-4-argon-retaking-benchmark-lead-over-openai-and-anthropic-but-in-limited-release)). **3. Cybersecurity defense is the headline capability.** Argon was trained to "autonomously find, validate, and patch critical software vulnerabilities." For trusted defenders and internal Google teams, it ships **without cyber guardrails**. Wiz is already using it via its "Scan for Good" initiative and says Argon uncovered a critical vulnerability exposing sensitive personal info in healthcare software used by hospitals worldwide — a risk previous frontier models missed ([Google blog](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon); [TechCrunch](https://techcrunch.com/2026/09/30/google-releases-gemini-4-argon-called-its-most-powerful-model-yet/)). **4. Real internal Google deployments.** Argon agents are migrating C/C++ codebases to Rust (from tens of thousands of lines up to 800K+ lines for the Fuchsia Zircon kernel); for libgav1 they replaced 32K lines of SIMD code, producing a memory-safe decoder **2.7x faster** than the prior Rust port. It also beat a published quantum-optimization baseline by **40%**, and autonomously freed **300+ TiB of data-center memory** (est. 500 TiB–1 PiB total) ([Google blog](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon)). **5. Pricing and phased rollout.** Introductory price: **$2 per million input tokens, $10 per million output tokens**, with cached input at 95% off. It is **not broadly available yet** — rollout begins with trusted cyber defenders via the **Fairwind Program**, while Google participates in the U.S. government's voluntary pre-release model-access process. Broader availability is planned "as soon as possible," starting with paid API customers and Google AI Ultra subscribers ([Google blog](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon); [VentureBeat](https://venturebeat.com/technology/google-unveils-gemini-4-argon-retaking-benchmark-lead-over-openai-and-anthropic-but-in-limited-release)). ## Why it matters - **Google is reasserting frontier leadership.** After months of pressure from OpenAI (GPT-6 Astra) and Anthropic (Fable/Opus), Argon gives Google its strongest claim yet to overall frontier leadership *by breadth* of benchmark wins — though the race stays close and remains workload-dependent ([VentureBeat](https://venturebeat.com/technology/google-unveils-gemini-4-argon-retaking-benchmark-lead-over-openai-and-anthropic-but-in-limited-release); [TechCrunch](https://techcrunch.com/2026/09/30/google-releases-gemini-4-argon-called-its-most-powerful-model-yet/)). - **It targets where enterprises already spend:** software development, professional knowledge work (legal/finance), and cybersecurity operations — not just chatbot demos. - **Safety-first, government-coordinated release.** The phased rollout follows CEO Sundar Pichai signing a voluntary AI safety accord with President Trump and major tech leaders at the White House the day before. Google says it is scaling safeguards in four cybersecurity areas (including misuse and prompt injection) before public launch. Tulsee Doshi, Google's Gemini model product lead, told CNBC the approach "gives us more confidence, but also enables us to put a model that is trained and strong in cyber defense in the hands of defenders as soon as possible" ([CNBC](https://www.cnbc.com/2026/09/30/google-gemini-4-argon-ai.html)). - **Timing:** Argon lands nearly a year after Gemini 3 put Google back at the forefront, and follows Google's recent pivot toward faster, lower-cost "flash" models ([CNBC](https://www.cnbc.com/2026/09/30/google-gemini-4-argon-ai.html)). ## Strongest sources - [Google — "Gemini 4 Argon: our next era of frontier intelligence" (primary announcement)](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon) - [VentureBeat — benchmark-by-benchmark breakdown vs. GPT-6 Astra and Claude Opus 5.5](https://venturebeat.com/technology/google-unveils-gemini-4-argon-retaking-benchmark-lead-over-openai-and-anthropic-but-in-limited-release) - [CNBC — rollout, pricing context, and the White House safety accord](https://www.cnbc.com/2026/09/30/google-gemini-4-argon-ai.html) - [TechCrunch — cybersecurity focus and competitive framing](https://techcrunch.com/2026/09/30/google-releases-gemini-4-argon-called-its-most-powerful-model-yet/) *Note: benchmark figures and the "industry-leading" claims are Google's own, as reported by the outlets above; independent third-party verification of the full benchmark set was not available in the sources opened.*