Google released Gemini 4 Argon on September 30, 2026, positioning the new model as its most capable AI to date. The launch marks a shift in the company’s roadmap, as Google confirmed it has canceled plans for Gemini 3.5 Pro to move directly from Gemini 3.1 Pro Preview to the Argon architecture.
The defining technical feature of Gemini 4 Argon is its 1-million-token output limit. While large context windows for data input have become standard, a 1-million-token output capacity is a significant increase from the 64,000-token limit found in previous iterations. This allows the model to engage in “long-horizon” tasks, such as generating massive codebases or conducting exhaustive security audits in a single uninterrupted response.
The Fairwind Program and Cybersecurity
Initial access to the model is being prioritized for the security sector. Through the Fairwind Program, Google is restricting early availability to vetted cybersecurity defenders. This rollout strategy focuses on using Argon’s reasoning capabilities to identify vulnerabilities and automate complex defensive maneuvers.
According to Help Net Security, Google has already deployed Argon internally to handle large-scale engineering transitions. The model was used to rewrite approximately 800,000 lines of the Fuchsia Zircon kernel from C++ into Rust, a move intended to improve memory safety. Additionally, Google reports that Argon helped optimize data center operations, freeing up 300 TiB of memory through more efficient resource allocation.
Benchmarks and Performance
Gemini 4 Argon’s performance puts it at the top of several industry intelligence indices. On the DeepSWE v1.1 benchmark, which measures an AI’s ability to solve real-world software engineering issues, Argon scored 77.9%. This result places it ahead of OpenAI’s GPT-6 Astra and Claude Opus 5.5 in similar evaluations. It also recorded a 51.3% on AutomationBench, a metric for autonomous task completion.
Pricing and Availability
Google is offering Gemini 4 Argon with an introductory pricing structure to encourage adoption among enterprise developers. The current rates are significantly lower than the projected standard costs:
- Introductory Rate: $2 per million input tokens / $10 per million output tokens.
- Standard Rate (Future): $4 per million input tokens / $20 per million output tokens.
While the model is currently focused on the Fairwind Program and internal Google projects, it is expected to expand to the broader Gemini Advanced subscriber base and general API users following the initial vetting period.
