Quick answer: On September 30, 2026, Google announced Gemini 4 Argon, its new frontier AI model for software engineering, enterprise knowledge work and cybersecurity. Its standout feature is a 1 million token output limit, up from 64,000. Access is limited at first to vetted cyber defenders through Google’s Fairwind Program, with paid API access and Google AI Ultra to follow.
What happened
Google DeepMind introduced Gemini 4 Argon as its “next era of frontier intelligence.” Rather than a broad consumer launch, Google is starting with security professionals because of the model’s ability to autonomously find and help fix software vulnerabilities. Google cited “the compute requirements and strict safeguards” a model with this level of autonomy needs.
Capabilities and benchmarks
- Output limit: 1 million tokens, up from 64K, enabling very long multi-step tasks
- DeepSWE v1.1: 77.9%, a state-of-the-art software engineering result per Google
- CWE-bench v1: 68%, tied for first in vulnerability detection
- LVBench: 91.7% for long-form video understanding
- Vals Index: Leading results across finance, coding, legal and tax tasks
Google also shared internal results: Argon helped migrate more than 800,000 lines of Fuchsia Zircon kernel code from C/C++ to Rust, made the libgav1 video decoder 2.7 times faster, found memory optimizations that freed over 300 terabytes across Google’s servers, and identified a severe vulnerability in hospital software that earlier models missed.
Pricing and rollout
Google’s announcement lists introductory API pricing of $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached input, rising later to $4 and $20. The rollout order is Fairwind Program cyber defenders first, then paid API tiers for developers and enterprises, then Google AI Ultra subscribers. Free Gemini users do not have access yet.
Why it matters
Argon raises the bar in Google’s race with OpenAI and Anthropic, which both shipped cheaper models in September. For SEOs and marketers, Gemini models power AI Overviews and AI Mode, so advances at the frontier tend to flow into Google Search over time. The limited, security-first rollout also signals that labs are now gating their most capable models behind safety programs.
Key facts
- Announced: September 30, 2026
- Developer: Google DeepMind
- Initial access: Fairwind Program (vetted cyber defenders)
- Output limit: 1 million tokens
- Intro API price: $2 input / $10 output per million tokens
FAQ
Can I use Gemini 4 Argon today?
Only if you are part of Google’s Fairwind Program. Paid API and AI Ultra access are planned next, with no firm dates yet.
Is Gemini 4 Argon used in Google Search?
Google has not said Argon powers AI Overviews or AI Mode. Search currently uses Gemini Flash models.
