Google DeepMind Unveils Gemini 4 Argon With 1M Output Tokens

Google DeepMind announced Gemini 4 Argon on March 18, 2026, introducing its new frontier model and the first in the Gemini 4 generation. This advanced AI is engineered for complex, long-horizon tasks, specifically targeting software engineering, enterprise knowledge work within sectors like legal and finance, and cybersecurity defense. The most significant technical advancement in Gemini 4 Argon is its vastly expanded output capacity, capable of generating up to 1 million tokens in a single response. This represents a substantial increase from the 64,000 output tokens available in previous Gemini models.

Google DeepMind has outlined a phased release strategy for Gemini 4 Argon. The company is actively participating in the U.S. government's voluntary process for pre-release model access. This approach will allow Google to gather crucial feedback from early testers and refine the model's guardrails before a broader public release. The pricing structure for Gemini 4 Argon has been made public, with an introductory offer of $2 per 1 million input tokens and $10 per 1 million output tokens. A significant discount of 95% is applied to cached input tokens, effectively reducing the cost to $0.10 per 1 million tokens. Following this introductory period, the pricing is set to increase to $4 per 1 million input tokens and $20 per 1 million output tokens. Logan Kilpatrick confirmed these introductory pricing details.

The capability to generate 1 million output tokens in a single response is a pivotal development, as current frontier AI APIs have much lower output limits. For comparison, models such as Claude Opus 5.5, Claude Fable 5.1, and GPT-6 Astra each support a maximum of 128,000 output tokens. Google's team emphasizes that Argon's ability to "think deeply" and generate hundreds of thousands of tokens in one continuous trajectory will empower developers to perform extensive code refactoring or generate lengthy reports without the need to break down tasks across multiple interactions. However, this enhanced capability comes with a notable cost, with a full 1 million output tokens costing $10 at introductory rates and $20 thereafter. Google has not yet disclosed the specific input context window size for Gemini 4 Argon.

In performance comparisons, Google DeepMind benchmarked Gemini 4 Argon against leading models including GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1. Argon demonstrated superior performance, leading outright on 12 out of 18 evaluated benchmarks and achieving a tie for first place on one benchmark. Notably, Argon achieved a state-of-the-art score of 77.9% on the DeepSWE v1.1 benchmark for long-horizon software engineering tasks. This performance surpasses that of Claude Opus 5.5, which scored 74.2%, and GPT-6 Astra, which achieved 74.1%. Argon also leads on the Vals Index, which measures economic impact across various sectors, although specific figures for this benchmark were not fully detailed in the announcement.