TL;DR

Google released Gemini 4 Hydrogen on 12 November, its quicker and cheaper sibling to the Carbon and Argon frontier models. It is meant for everyday coding and agent tasks, either alone or as the worker model under a larger one. Hydrogen is rolling out worldwide in the Gemini API, AI Studio, Gemini for business, the Gemini app and Google Search.

Input costs $1.50 per million tokens and output $7.50 per million, at the same rate across the whole 1 million-token context window. The launch follows our report on Gemini 4 Carbon’s wider release.

What Google claims

The benchmark figures are Google’s own. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default “high” thinking level against Argon’s 50, and 54 at a new “xhigh” setting. Compared with Gemini 3.8 Flash, the model Google expects teams to migrate from, it says Hydrogen completes tasks 2.5 times faster, more than triples its Terminal-Bench 4.1 score (57% against 18.2%) and produces over five times fewer fabricated claims.

Google says partners testing it on code and data-heavy enterprise work have seen runs 2x to 2.8x quicker than 3.8 Flash, using 12% to 23% fewer tokens. Hydrogen accepts up to 1 million tokens in and can produce as many out, and offers four thinking levels so users pay only for the reasoning a task needs.

Model pairing and safety

Alongside the model, Google added Orchestrator to its Antigravity coding environment. It moves work between models within one session, for example Carbon for planning and Hydrogen for writing and testing code. On Terminal-Bench 4.0, Google says this came close to Carbon’s success rate while using 24% less quota and finishing 19% faster.

On safety, Google says Hydrogen reduces misaligned actions and shortcuts twelvefold in tool-use and coding evaluations, and that it is more than 2.3 times as resilient to indirect prompt injection on the Gray Swan benchmark. Its false refusal rate on HealthBench Pro is under 0.4%, according to the company. A model card gives more detail.

Looking Forward

For UK businesses, the practical change is cost per task rather than peak capability. In our view, Google is pricing Hydrogen for the execution role in that pairing, where an expensive model plans and a cheap one does the bulk of the work. The gains are vendor-reported and largely against Google’s own older model; independent comparisons with rival small models will matter more than the launch claims.