Gemini 4 Argon Launches Globally with 1M-Token Context Window

Google DeepMind released Gemini 4 Argon to global users on October 13, 2026. The frontier AI model had first reached trusted participants in the Fairwind Program on September 30 before wider access opened through the Gemini API, Google Cloud, and the Gemini app.

Gemini 4 Argon Technical Capabilities and Performance

The model uses a unified 1,048,576-token context window shared across input, reasoning, and output. It supports long decode runs up to one million tokens when executed in the background through the Interactions API, while synchronous responses default to roughly 64,000 tokens. Users can select low, medium, or high thinking levels to balance speed and depth.

On benchmarks, Gemini 4 Argon reached 77.9 percent on DeepSWE v1.1 for long-horizon software engineering and 68 percent on CWE-bench v1 for vulnerability remediation. It also posted 91.7 percent on LVBench long-video understanding and ranked first on the Vals Index measuring economic impact across finance, coding, legal, and tax domains. Internal teams used the model to refactor six million lines of Chromium code and to migrate more than 800,000 lines of C and C++ to Rust in the Fuchsia Zircon kernel, producing memory savings above 300 TiB and a 2.7-times speedup in video decoding.

The model handles agentic coding and multi-step enterprise workflows in legal, finance, and scientific domains. Output style is described as more direct with reduced hedging and repetition compared with earlier Gemini versions.

Cybersecurity Focus, Pricing, and Rollout Timeline

Gemini 4 Argon was trained for autonomous vulnerability discovery, validation, and patching. Participants in the Fairwind Program and select internal teams received versions without additional cyber guardrails. Early deployment by Wiz through its Scan for Good initiative identified a critical flaw in healthcare software that earlier models had missed, protecting hundreds of thousands of patient records. The model scored 0.7 percent on the Gray Swan IPI prompt-injection benchmark.

Introductory pricing through November 30, 2026, stands at two dollars per million input tokens and ten dollars per million output tokens. Cached inputs receive a 95 percent discount. After that date, rates rise to four and twenty dollars respectively. Access is available under the model identifier gemini-4-argon in AI Studio, Google Cloud Vertex AI, and the Gemini app for Pro and Ultra subscribers. Free and standard Plus users remain limited to lighter models.

Google followed a phased release that included U.S. voluntary pre-release review and iterative safety improvements. Applications remain open for additional researchers on the free Antigravity tier and for more cyber-defense organizations through the Fairwind Program. Further model iterations are already in internal testing.