Gemini 4 Argon: 1M Tokens, Coding, and Cyber Defense

⚡ The Gist in 5 Seconds - The point: Google has unveiled the frontier model Gemini 4 Argon for complex, long-running tasks in software development, legal and financial work, and cyber defense. - Availability: the model is being rolled out to trusted cyber defenders through the Fairwind program and is already in use inside Google. - Limitation: due to its high level of capability, the release is staged: first through participation in the US government's voluntary pre-release access process, with a public release later. ### 🔍 What Was Found In a September 30, 2026 post, Koray Kavukcuoglu (SVP at Google DeepMind and Google's Chief AI Architect) announces the launch of Gemini 4 Argon. The model is already running inside the company: thousands of employees use it for specialized coding, research, and writing tasks. Internal results are impressive: in quantum algorithm optimization, Argon improved a baseline metric by 40% in minutes; the model's agents analyzed telemetry and applied memory optimizations, freeing over 300 TiB of memory in data centers, with total savings estimated at 500 TiB to 1 PiB. Migration of C/C++ codebases to Rust is also underway: from tens of thousands of lines in the re2 and libgav1 libraries to 800K+ lines in the Fuchsia Zircon kernel. A key feature of Argon is its expanded output token limit: 1 million versus the previous 64 thousand. This allows the model to generate hundreds of thousands of tokens in a single pass, deepening its reasoning. Launch pricing: $2 per million input tokens, $10 per million output tokens, with cached inputs at a 95% discount. The model sets new records: DeepSWE v1.1 — 77.9% on real-world software engineering tasks, leadership in the Vals Index (finance, coding, law, taxes), 51.3% on AutomationBench Zapier, and 91.7% on LVBench for long videos. In cybersecurity, Argon can autonomously find, verify, and fix critical vulnerabilities. The company Wiz is already using it in the Scan for Good initiative, and the model discovered a critical vulnerability exposing sensitive data. ### 💡 Why It Matters Expanding the context to 1M tokens means AI can process enormous amounts of information in one go — from analyzing large codebases to legal documents and financial research. Autonomous cyber defense with vulnerability patching could dramatically speed up threat response, especially for critical infrastructure. Google's internal examples show practical benefits for engineering productivity, and the benchmarks confirm superiority in economically significant workflows. ### 🧩 Context Google emphasizes that safely releasing such capabilities requires a staged approach. Argon is already part of the US government's voluntary pre-release access process, and the company is gathering feedback from early testers, iteratively improving safeguards ahead of a broad launch for developers, enterprises, and consumers. In parallel, the model is actively used inside Google, where it accelerates various areas, including quantum computing, opt

🔗 Read on blog.google