Google launches Gemini 4 Hydrogen for agentic workflows

Google introduced Gemini 4 Hydrogen on Thursday, a model built for fast, efficient execution across agentic workflows and everyday coding, positioned alongside its more deliberate frontier siblings, Gemini 4 Carbon and Gemini 4 Argon, which the company reserves for deep, long-horizon reasoning. Hydrogen carries one flat rate across its full 1-million-token context window — $1.50 for every million input tokens, $7.50 for every million output tokens. On the Artificial Analysis Intelligence Index, Google says Hydrogen already edges past Argon at the default high-thinking tier — 51 points to Argon’s 50 — climbing to 54 once pushed to the new xhigh setting, while often working faster than other models in its class.

Google is also introducing Orchestrator in Google Antigravity, which switches between models such as Carbon and Hydrogen within a single session instead of relying on one throughout a task; the company reports that on Terminal-Bench 4.0, Orchestrator’s task-success rate nearly matched Carbon’s while using 24% less quota on average and finishing about 19% faster. Against Gemini 3.8 Flash, the model Hydrogen is meant to replace, Google claims a 2.5-times speed-up at the default setting, a jump in Terminal-Bench 4.1 scores from 18.2% to 57%, and fabricated claims cut by more than five-fold, alongside competitive results across coding, agentic-research and multimodal-document benchmarks it tracks internally and against third-party tests.

On safety, Google reports steep drops versus 3.8 Flash on two measures it tracks closely — task overreach down as much as 81%, under-service down as much as 79% — plus a 12-fold cut in misaligned or shortcut-taking actions across its tool-use and coding evaluations, odds of bypassing a permission boundary on blocked tasks cut roughly in half, and a 96.7% score on its internal agent-honesty check; resistance to indirect prompt-injection attacks on the Gray Swan benchmark reportedly more than doubled. Hydrogen offers four thinking levels — low, medium, high (the default) and a new xhigh tier — letting users trade cost and latency for reasoning depth.

The model is rolling out today across Google’s developer, enterprise and consumer surfaces: the Gemini API and AI Studio, the paid tier of Android Studio, Gemini for business, Google Antigravity for Google AI Pro and Ultra subscribers, Gemini in Google Sheets, the Gemini app’s model picker, and Google Search.

Why it matters

By giving its fastest, cheapest Gemini 4 tier the same global, same-day rollout as a flagship — into Search, the Gemini app, Antigravity and the API all at once — Google is betting that most day-to-day AI work, not frontier reasoning, is where the bulk of real-world usage will concentrate.