Google unveils Gemini 4 Argon, still locked to internal testers

Google had promised Gemini 3.5 Pro in June, but it spent the summer releasing smaller Flash models instead. The arrival of Gemini 4 Argon marks a return to the frontier for the company, which says its engineers are already using the model internally for tasks such as migrating codebases to Rust and saving memory across data centers. Google has also provided benchmark results, including a 77.9 percent score on the DeepSWE v1.1 software engineering test, which it says beats competing models from other labs.

For now, the model remains in limited testing, and Google has not announced API pricing. The company has confirmed one concrete improvement for future users: a 1 million token output limit, up from 64,000 tokens in prior Gemini models, which Google says will let users complete more daunting tasks in a single step.

The stakes hinge on whether outside developers and businesses will be able to test the model and whether Google's performance claims hold up under independent scrutiny. Until pricing and broader access are announced, the competitive picture remains based on Google's own numbers.