AI
Google launches Gemini 4 Hydrogen, a faster Flash replacement with sharper reasoning
The tech giant's new model trades the Flash branding for an elemental naming scheme while delivering significant speed and accuracy improvements for most users.
Ars Technica
Google has introduced Gemini 4 Hydrogen, marking a generational step forward in its core AI model lineup. The new release ditches the "Flash" branding in favor of the company's emerging elemental naming convention—a shift that signals a broader pivot in how Google categorizes its AI offerings.
What Hydrogen replaces#
Hydrogen is designed as the direct successor to Gemini 3.8 Flash, the workhorse model powering most of Google's consumer-facing AI products. Unlike the company's more recent flagship releases—Argon and Carbon—which faced staged rollouts restricted initially to researchers and paying subscribers, Hydrogen rolled out with minimal gatekeeping. The model is now live across Google's product ecosystem.
Performance gains#
The new model delivers measurable leaps in speed and accuracy. In its highest reasoning mode (xhigh), Hydrogen runs 2.5 times faster than its predecessor while maintaining or improving quality. Google reports a more than fivefold reduction in hallucinations compared to Flash 3.8. For developers, the improvements are even sharper: code generation performance more than triples on standard benchmarks like Terminal-Bench.
In head-to-head testing against the pricier Argon model, Hydrogen proves competitive—achieving a score of 54 in the Artificial Analysis Intelligence Index when using its top reasoning tier, just shy of Argon's 50.
How it works#
Hydrogen operates across four distinct thinking levels (low, medium, high, and xhigh), allowing users to dial in the balance between speed and depth of reasoning based on their needs. This flexibility lets both casual users and power users tune performance to match their workload.

