Google launches Gemini 4 Hydrogen, its new consumer AI model

Google has launched Gemini 4 Hydrogen, an artificial intelligence model designed to replace Gemini 3.8 Flash across all its products. Available immediately and without subscription restrictions, it offers significant gains in speed and reliability.

Google continues its cycle of renewing its artificial intelligence models with Gemini 4 Hydrogen, now deployed on a large scale. Unlike the company’s recent releases, notably Argon and Carbon, whose launch was initially reserved for cybersecurity researchers and then for subscribers to the most expensive plans, Hydrogen is accessible without delay or special subscription conditions.

This model replaces Gemini 3.8 Flash, which until now was the reference for common uses within the Google ecosystem. The company claims substantial improvements: in high reasoning mode, Hydrogen is said to be 2.5 times faster than its predecessor. The model operates according to four processing levels, from the lightest to the most intensive, the latter level called “xhigh” being a novelty of this generation.

In terms of reliability, Google indicates that false or fabricated statements by the model have been reduced by a factor greater than five compared to 3.8 Flash. Code generation also shows marked progress: Hydrogen would multiply by more than three the score obtained by 3.8 Flash in the Terminal-Bench benchmark, used to evaluate the programming capabilities of models.

Hydrogen’s performance places it in an unexpected position relative to Argon, a higher-end model launched earlier this year. In the Artificial Analysis Intelligence Index benchmark, Hydrogen in xhigh mode scores 54, compared to 50 for Argon, suggesting that the gap between the two ranges has significantly narrowed.

Google recently adopted a nomenclature based on chemical elements for its AI models, abandoning the Pro and Flash designations that more clearly indicated each version’s positioning. Hydrogen fits into this new naming logic while explicitly targeting users seeking speed and resource efficiency.