Gemini 4 Hydrogen is Google’s model for everyday tasks, coding, and agentic work, with text, image, video, and speech input, text output, a 1M-token context window, and low, medium, high, and xhigh reasoning levels. Google says it is rolling out across Google AI Studio, the Gemini API, Gemini, Search, and Antigravity, per X, according to @Google.
Artificial Analysis reports an Intelligence Index score of 54 at xhigh, one point below Claude Sonnet 5.5 and GPT-6.1 Sol at max reasoning, and four points above Gemini 4 Argon at high. At high, Hydrogen scores 51; the post compares that with 52 for GPT-6.1 Sol at xhigh and Claude Opus 5.5 at high, according to @ArtificialAnlys.
The reported strengths are visual reasoning and avoiding fabricated answers: Hydrogen scores 33% all-pass on the long-document visual benchmark GDP.pdf at medium reasoning, tied for first, and has an 8% hallucination rate on AA-Omniscience, compared with 43% for Claude Sonnet 5.5 and 52% for GPT-6.1 Sol. Artificial Analysis also says its presentation on AA-Briefcase rates above other Gemini models, including Carbon, according to @ArtificialAnlys.
On Terminal-Bench 4.0, Hydrogen scores 54%, below Claude Sonnet 5.5’s 64% and GPT-6.1 Sol’s 56%. On Terminal-Bench-Science 0.1 it scores 50%, versus 53% and 58%, respectively. Xhigh raises Hydrogen’s science score from 41% at high, but leaves its Terminal-Bench 4.0 score unchanged at 54%, according to @ArtificialAnlys.
Hydrogen is priced at $1.50 per million input tokens and $7.50 per million output tokens; cached input is $0.075 per million. Artificial Analysis estimates its xhigh cost per task at about 32% of Sonnet 5.5’s and about 2.5 times GPT-6.1 Sol’s. It reports $3.16 per task at xhigh versus $1.87 at high, and says xhigh nearly doubles output tokens for a three-point Intelligence Index gain; medium and low cost per task are reported as $0.80 and $0.29, with scores of 41 and 27, according to @ArtificialAnlys.
Artificial Analysis reports output speed of about 200 tokens per second and places Hydrogen with high, medium, and low reasoning on its time-per-task versus intelligence Pareto frontier. Google separately claims Hydrogen finishes tasks 2.5 times faster than 3.8 Flash; that is a Google-reported comparison, not a speed comparison with Claude Sonnet 5.5 or GPT-6.1 Sol, per X, according to @ArtificialAnlys.