
Hydrogen replaces Gemini 3.8 as the default model for Gemini and Search and will likely be used by billions of people in short order.
The new model leaps ahead of Gemini 4 Argon on its default settings at Artificial Analysis’ benchmark of benchmarks, but more notably it ranks seventh on time per task, just behind GPT-6 Luna and Mistral Large.
It is also 3.5 times faster than the Gemini 3.8 Flash it is replacing, while more than tripling its Terminal-Bench score.
Google still recommends using Carbon or Argon for the more demanding tasks, while having Hydrogen on hand for simpler, everyday tasks.
To make better use of their models’ expanding availability, Google is also introducing Orchestrator in Google Antigravity that switches between models based on task complexity. The feature completes tasks 19% faster using 24% less quota. Hydrogen can also act as a worker agent with Carbon coordinating.
Read more: Google’s presentation, launch post. The Verge, Ars Technica and TechCrunch. Discussion on Hacker News and r/Singularity.