Google announced Gemini 4 Argon on Tuesday, its newest frontier model built for long-horizon professional work in software engineering, enterprise knowledge tasks, and cybersecurity defense. Writing on the official Google blog, Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect, said the model is rolling out initially to a set of trusted cyber defenders through Google's Fairwind Program, with broader access to follow once guardrails are strengthened.
The announcement lands during one of the busiest stretches of the year for frontier AI releases. For continuous coverage of model launches, policy fights, and funding rounds, AI Buzz Wire tracks the developments that matter as they happen.
A phased rollout shaped by caution
Unlike a typical major model launch, most users cannot try Argon today. Google said it is engaged in the U.S. government's voluntary process for pre-release model access and will gradually expand availability as feedback from early testers shapes its guardrails. Reuters reported the flagship arrives after months of delays, and VentureBeat noted the limited-release strategy still allows Google to claim a retaken benchmark lead over OpenAI and Anthropic. Bloomberg reported measurable skepticism about the new model among Google's own employees ahead of launch.
When wider access comes, Google said it will start with paid API customers and Google AI Ultra subscribers.
What Google says Argon can do
The headline capability change is endurance. Argon's output token limit has been expanded to an industry-leading 1 million tokens, up from the previous 64K. Google argues that giving a model the headroom to reason across hundreds of thousands of tokens in a single trajectory lets it solve hard problems in one pass rather than fragmenting work across sessions.
Google published benchmark results to back the frontier claim:
- DeepSWE v1.1: 77.9% — a new state of the art on the real-world software engineering benchmark
- Vals Index: No. 1 — leading performance across finance, coding, legal, and tax work weighted by contribution to U.S. GDP
- AutomationBench: 51.3% — first place on Zapier's end-to-end business execution benchmark
- LVBench: 91.7% — state of the art in long-video understanding
- CWE-bench v1: 68% — tied for first on security vulnerability remediation
Inside Google's own workflows
Argon is already powering internal work at Google, and the company offered unusually concrete examples. On quantum computing, the model optimized spacetime resources for bottleneck subroutines, beating a published baseline by 40% in minutes. Analyzing fleet-wide profiling telemetry, teams of Argon agents identified and applied memory optimizations across Google's data centers that freed over 300 TiB once rolled out, with an estimated 500 TiB to 1 PiB in total savings.
The model is also driving large-scale code migrations from C/C++ to Rust, scaling from tens of thousands of lines in core libraries like re2 and libgav1 up to more than 800,000 lines in the Fuchsia Zircon kernel. For libgav1, Google's open-source video decoder, Argon agents replaced 32,000 lines of SIMD code with compiler-friendly safe Rust, producing a memory-safe decoder that runs 2.7 times faster than the previous Rust port with identical output.
Cybersecurity defense comes first
Argon was explicitly trained for defensive cyber work: autonomously finding, validating, and patching critical software vulnerabilities. For trusted defenders and Google's internal teams, the company will release the model without cyber guardrails so defenders get full capability. On CWE-bench v1, which measures vulnerability remediation, Argon ties for first place at 68%, building on the frontier performance of 3.8 Flash Cyber, and Google says Argon outperformed that model in attack-surface discovery and proof-of-concept generation on Wiz's internal black-box penetration testing benchmark.
Security firm Wiz is already using Argon through its Scan for Good initiative, which protects critical public infrastructure for free. In an early demonstration, Google says the model uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide — a risk previous frontier models had missed.
Four layers of frontier safeguards
Before broad availability, Google is strengthening safeguards in four areas. Against misuse, Argon is designed to refuse cyber and CBRN attack requests while preserving legitimate dual-use research, with new techniques that monitor the model's internal activations to spot abuse. Against prompt injection, adversarial training has made Argon Google's most resilient model yet, leading on Gray Swan's Indirect Prompt Injection benchmark.
For misalignment, Google is deploying mitigations that monitor Argon's chain of thought and actions, halting execution when necessary, with alerts routed to a dedicated incident response team. The company also hardened its sandboxed training and evaluation environments, sealing them before high-risk runs. Notably, Google urged the rest of the industry to preserve reasoning transparency so model thoughts remain useful for diagnosing misalignment.
Pricing and availability
Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off the input price. After the introductory period, pricing rises to $4 per million input tokens and $20 per million output tokens — positioning Argon aggressively against rival frontier offerings.
The staged debut reflects a new reality for frontier releases: capability is now table stakes, and the differentiator is demonstrating control. Google is betting that cyber defenders, hospital software, and Rust migrations are the right showcase — and that a slower public rollout costs less than another safety incident. Developers and enterprises should watch for the paid API and AI Ultra access window, which Google says will open as soon as guardrail testing allows.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
