AI/ google · gemini · ai-models · benchmarks

Gemini 4 Hydrogen Replaces 3.8 Flash Across Google Products

Google says its new Flash replacement runs 2.5 times faster, and unlike its Argon and Carbon models, access starts immediately.

Google has replaced Gemini 3.8 Flash with Gemini 4 Hydrogen across its products, and unlike the last two big releases, nobody has to wait or pay top dollar to try it.

Hydrogen takes over the faster, cheaper tier that most people in Google's ecosystem actually use. It runs at one of four thinking levels (how much reasoning the model does before it answers): low, medium, high and a new xhigh. Google says that even in high mode, Hydrogen is 2.5 times faster than 3.8 Flash. It also says Hydrogen cuts "fabricated claims" by more than 5x, according to its own analysis, and more than triples 3.8 Flash's score on Terminal-Bench, a code generation benchmark.

The most eye-catching number is the comparison with Argon, the larger model above it. In xhigh mode, Hydrogen scored 54 on the Artificial Analysis Intelligence Index, against 50 for Argon.

The rollout differs sharply from Argon and Carbon. Those began with cybersecurity researchers, and even at general availability they were limited to subscribers of Google's most expensive AI plan. Hydrogen's limits are much more relaxed, and access begins immediately.

That matters more than the benchmark. Most people never pick a model by name. They get whatever sits behind the button, and for most of Google's users that is now Hydrogen. The elemental names say nothing about size or speed, which the old Pro and Flash labels at least hinted at.

One caution on the numbers: the 2.5x speedup is quoted for high mode, while the 54 comes from xhigh. They may not describe the same experience, and one index is a thin basis for saying a cheaper model has caught its bigger sibling. If Hydrogen really does trade blows with Argon, the open question is what the premium tier is for.

TR

The Revision

Written by an AI system from the public sources credited above. How we write →