Google released Gemini 4 Hydrogen on November 12 and immediately put it somewhere billions of people would notice: Search.

What the model is. Hydrogen is the fast, cheap member of the Gemini 4 family, replacing Gemini 3.8 Flash as the default for high-volume work. Google says it finishes tasks about 2.5 times faster than its predecessor, offers four thinking levels so developers can trade latency for reasoning, and handles a one-million-token context window with up to a million tokens of output. Pricing is $1.50 per million input tokens and $7.50 per million output, flat across the whole context window, with cached input at roughly a twentieth of that.

The benchmark claims are unusually specific. Google says Hydrogen scored 51 on the Artificial Analysis Intelligence Index at its default thinking level, ahead of the larger Gemini 4 Argon at 50, rising to 54 at the new top setting. On Terminal-Bench 4.1 it reports 57%, against 18.2% for the model it replaces. Google also claims fabricated claims drop by more than five times, which is the number that matters most for anyone pointing this at documents.

These are vendor-selected results and should be read as such. The independent part is the Artificial Analysis score, and even that is a composite that weights tasks the index's authors chose.

What changes in Search. Robby Stein, Google's vice president of product for Search, called Hydrogen the biggest model upgrade to Search yet. It is now the default in AI Mode and handles the hardest questions inside AI Overviews. Google says it copes better with long, multi-part questions, with typos, and with requests that need several answers; that responses get more visual, with comparison tables and generated tools; and that an upgraded query fan-out reaches a wider set of websites. Answers improve across all the languages AI Mode supports.

That last point is both a promise and a warning for publishers. Query fan-out means one user question becomes many background searches. A wider set of linked sources is good for small sites. A model that answers the question directly is not. The referral economics of the open web are being renegotiated by every one of these releases, and Hydrogen's efficiency makes the negotiation cheaper for Google to run at scale.

Who actually gets it. Google AI Plus at $4.99 lists Hydrogen as its headline model, AI Pro at $19.99 includes Hydrogen and Argon, and AI Ultra from $99.99 has the highest limits across all three. Free users cannot choose it: they get it through automatic model selection when Google decides it fits the prompt, and reports suggest most free prompts still go to the lighter Flash-Lite model. That gap is where the real product tiering lives.

For developers, the more interesting release is Orchestrator mode in Antigravity, which routes work across Gemini models automatically: Carbon for planning, Hydrogen for writing code and running tests. Google says that combination finishes complex tasks 19% faster while consuming 24% less quota than using Carbon alone.

What to watch. Whether Hydrogen holds up on long-context work where cheap models usually degrade, whether the introductory pricing holds, and whether Search users can tell the difference in anything other than speed. Speed, in consumer AI, is not a detail. It is most of the experience.