Google Launches Gemini 4 Hydrogen for Agentic Workflows

Google has launched Gemini 4 Hydrogen, a faster Gemini 4 model that DeepMind describes as “built for rapid execution across diverse agentic workflows.” The announcement, dated 12 November 2026, is by Koray Kavukcuoglu, senior vice president at Google DeepMind and Google’s chief AI architect. Gemini 4 Carbon and Gemini 4 Argon remain the frontier models for “deep, long-horizon reasoning.”

The company says Hydrogen is “rolling out globally” from 12 November across developer, enterprise and consumer products. It is in the Gemini API in Google AI Studio and on the paid tier in Android Studio, in Gemini for business, in Google Search for all users, in Gemini in Google Sheets starting with Google AI Pro and Ultra subscribers, and in Google Antigravity for paid AI Pro and Ultra subscribers. In the Gemini app it appears in the model drop-down; a footnote on the DeepMind post says free users will get it “via auto model selection when it’s the appropriate model for the prompt.” That is not the same as a free, manual choice of Hydrogen for every query.

A faster execution model, not a replacement for Carbon or Argon

Kavukcuoglu’s post places Hydrogen beside, rather than above, the two frontier models. Teams still need Carbon and Argon, it says, for deep architectural judgement and hard bugs; Hydrogen is meant to write code, run tests and verify changes more quickly, either on its own or as the execution engine in a multi-model workflow. A developer post published the same day repeats that it can run standalone across four thinking levels or alongside Carbon and Argon.

Those thinking levels are low, medium, high (the default) and a new “xhigh” setting. Google prices the Gemini API at $1.50 per million input tokens and $7.50 per million output tokens, flat across Hydrogen’s 1 million-token context window. The model can also emit up to 1 million output tokens. That price is what Google has published for Hydrogen; this article does not compare it with Carbon or Argon.

The same post introduces Orchestrator in Google Antigravity, which switches between models such as Carbon and Hydrogen in a single session. On Terminal-Bench 4.0, Google says Orchestrator reached “near-Carbon task success rates while, on average, consuming 24% less quota and completing 19% faster.” Carbon can act as a coordinator, it adds, while Hydrogen runs worker and verifier agents. Those figures, like the others below, are Google’s.

This is a third member of a family TSN has already been following: Carbon is generally available, after an earlier rollout to cyber defenders ahead of developers, and Argon has had a global release, including general availability in GitHub Copilot. Hydrogen’s developer path, as announced today, is Google’s own API, studios and Antigravity, not a Copilot listing.

Google’s own scores, and a Search upgrade

Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, against 50 for Argon, and 54 at xhigh. Versus Gemini 3.8 Flash, it says Hydrogen is “finishing tasks 2.5 times faster” at the high default setting, scores 57% on Terminal-Bench 4.1 against 18.2%, and is “cutting fabricated claims by more than 5x.” In early testing, Google says, partners saw the new model run 2x to 2.8x faster and use 12% to 23% fewer tokens than 3.8 Flash. Other scores Google lists include 58.8% on HLE-Diamond, first place on the Vals Tax Agent Bench at 52.3%, a tie for first on the Vals Legal Research Bench at 55.8%, fifth place on arena.ai’s Agent Arena, 53.4% on the Multimodal Knowledge Agent benchmark, and 71.6% on Agents’ Last Exam v1.1.

None of those results has been independently reproduced in the sources for this article. They should be read as Google’s claims.

In Search, Hydrogen is available to all users worldwide, according to Robby Stein, vice president of product for Google Search. “This is the biggest model upgrade to Search yet,” Stein says. Google says the model improves intent understanding, including for multi-part queries and typos, upgrades query fan-out so that results include a wider range of links, produces more visual responses such as tables and images, and offers richer language understanding across the languages supported in AI Mode.

Safety figures Google gives against 3.8 Flash

The DeepMind post also sets out safety and reliability figures, all compared by Google with Gemini 3.8 Flash or with internal tests. It says overreach in agentic tasks is down “by up to 81%,” under-service “by up to 79%,” and misaligned actions and shortcutting “by 12x.” It reports a 96.7% score on an internal agent honesty evaluation, says Hydrogen is “more than 2.3x more resilient” on the Gray Swan IPI benchmark for indirect prompt injection, and gives HealthBench Pro false refusals as “under 0.4%.” The post points readers to a Gemini 4 Hydrogen model card for more detail; that card has not been read for this article, and no figures are taken from it.

Google’s earlier safety report on Argon said it could not rule out critical cyber and biological risk thresholds. Today’s Hydrogen posts do not revisit that finding, and they should not be read as if they had closed it.

Google’s scores describe a faster teammate, not a settled ranking

Hydrogen is being sold as the quick execution layer in a three-model Gemini 4 stack, not as a quiet replacement for Carbon or Argon. That distinction matters when Google also says the new model edges Argon on the Artificial Analysis Intelligence Index at the default setting. A one-point lead on a vendor-cited index, at one thinking level, does not by itself change the roles Google has assigned. Carbon and Argon still do the “deep, long-horizon reasoning”; Hydrogen, in the company’s own words, is built for rapid agentic work.

Every headline number in the launch — the 2.5 times speed-up, the Terminal-Bench jump from 18.2% to 57%, the partner tests, the Orchestrator quota and time savings, and the safety percentages — comes from Google. There are no independent benchmarks in the material published today. Until other evaluators run the same tests, those figures are a first-party scorecard, useful as a statement of what the company claims, not as a confirmed ranking against rivals.

Availability is similarly more specific than the “everyone” language in the posts. Search users all get the upgrade. Developers on the Gemini API and AI Studio can build with it, with Android Studio limited to the paid tier and Antigravity limited to paid Pro and Ultra subscribers. Sheets starts with those same paid tiers. In the Gemini app, paying users can pick Hydrogen from the model drop-down; free users receive it only if automatic model selection decides it is the appropriate model for that prompt. That is a real expansion of Gemini 4, and it is not the same as putting Hydrogen, on demand, in every free chat.

The open questions are therefore practical. Do independent evaluations agree with Google’s speed, honesty and injection-resistance numbers? Does the Orchestrator actually keep near-Carbon success rates once people use it outside Terminal-Bench 4.0? And, for free app users, how often does automatic selection in fact choose Hydrogen? The posts do not answer those questions. What they do settle, as of 12 November, is that Gemini 4 now has a third, faster model, that Carbon and Argon remain the frontier reasoning pair, and that the first set of scores is Google’s own.

Sources

1. Koray Kavukcuoglu, Google DeepMind, “Introducing Gemini 4 Hydrogen,” 12 November 2026. https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/
2. Robby Stein, “Google Search gets a major upgrade with Gemini 4 Hydrogen,” 12 November 2026. https://blog.google/products-and-platforms/products/search/gemini-4-hydrogen-search-upgrade/
3. Google, “Gemini 4 Hydrogen: everyday coding and agentic workflows for developers,” 12 November 2026. https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-4-hydrogen/

Related stories

Gemini 4 Carbon Is Generally Available, Google Says is the nearest TSN account of the frontier model Google is now pairing with Hydrogen, and the reason Carbon remains in the picture.

Also relevant:

• Google Releases Gemini 4 Argon Worldwide: What It Costs, Who Gets It and What Google Claims
• Google Says It Cannot Rule Out Gemini 4 Argon Reaching Critical Cyber and Bio Risk Levels