Google introduced Gemini 4 Hydrogen, the third model in the Gemini 4 series. While Argon and Carbon are intended for long and complex tasks, Hydrogen is designed for the rapid execution of everyday programming and agent workflows.
Unlike the previous two, Hydrogen is also available on the free tier in Google AI Studio and the API. In Search, Google is rolling it out to all users globally, across all languages supported by AI Mode. According to the help page, in the Gemini app it can be selected by Google AI Plus, Pro, and Ultra subscribers, while free users get it only when the automatic selection determines that the prompt requires it, since most answers are still provided by Flash-Lite. In Antigravity, it is available to companies, and among regular users, only to AI Pro and Ultra subscribers.
The API pricing is $1.50 per million input tokens and $7.50 per million output tokens, remaining the same for the entire one-million token context window. This is less than Argon and Claude Sonnet 5.5, but more than GPT-6.1 Terra.Additionally, Gemini 3.8 Flash costs half as much until the end of the year, and starting in January, it will cost the same as Hydrogen.
According to Google, compared to 3.8 Flash, the new model at the base level completes tasks 2.5 times faster, scores 57% versus 18.2% on Terminal-Bench 4.1, and “hallucinates” over five times less. The model also features four thinking levels, of which “xhigh” is new.
In Google’s table, Hydrogen is compared to Sonnet 5.5 and GPT-6.1 Terra, ranking first in most tests. However, Sonnet 5.5 is ahead of it on Terminal-Bench 4.1 (63.9% versus 57%), on computer control in OSWorld-2.1 (83.9% versus 77.6%), and on the scientific Terminal-Bench-Science (59.9% versus 51.4%).
According to Google, on the Artificial Analysis Intelligence Index, Hydrogen scores 51 points at the base “high” level—one more than Argon—and 54 at the “xhigh” level. On Google’s chart, Sonnet 5.5 achieves the same result at its highest level but is several times more expensive per task, while on the ranking list itself, it is slightly ahead of Hydrogen with 55 points.
In Antigravity, Google is also introducing Orchestrator, which switches between Carbon and Hydrogen within the same session. According to the company, on Terminal-Bench 4.0 it nearly reaches Carbon’s success rate, while on average consuming 24% less of the limit and being 19% faster.
It appears that with Hydrogen, much like OpenAI with Terra, Google is not aiming for the absolute highest score, but rather for a balance of price and speed.


