 |
| AI-generated illustration of a glowing blue AI core orb at the centre of a search interface, with light beams fanning out into search result cards and a magnifying glass |
On Thursday, November 12, 2026, Google put a new artificial intelligence model at the centre of almost everything it ships. Gemini 4 Hydrogen, announced that day by Google DeepMind, is now rolling out across the Gemini app, Google Search, the Gemini API, Google AI Studio, Gemini for business, and Google Antigravity. It replaces Gemini 3.8 Flash, the fast model that powered most of Google's everyday AI features until now, and Google is describing it as the execution engine of the Gemini 4 family: the model that does the work, while its bigger siblings, Gemini 4 Carbon and Gemini 4 Argon, handle the deepest reasoning. The most striking claim comes from Robby Stein, Google's vice president of product for Search, who called Hydrogen "the biggest model upgrade to Search yet" and said it is now the default model in AI Mode and the model used for the hardest questions in AI Overviews. That means the model is not just a developer release. It is already answering questions for everyone who searches on Google.
The launch matters for a simple reason: speed is the part of AI that users actually feel. A frontier model that takes a minute to think is impressive in a demo and useless in a search box. Google's pitch for Hydrogen is that it closes that gap, finishing everyday agentic work much faster than the model it replaces while staying smart enough to beat more powerful models on at least one broad benchmark. Every performance number in this article comes from Google itself, and that is worth saying plainly, because no independent lab has yet reproduced these results. The launch is real, the rollout is under way, and the claims are Google's to prove over the coming weeks. What follows is what Google announced, what it costs, where you will meet the model, and the questions worth asking before you build anything on it.
As an Amazon Associate, we earn from qualifying purchases.
Prices and savings can change. Check the retailer for the current price.
What Gemini 4 Hydrogen is
Gemini 4 Hydrogen is the third model in the Gemini 4 family, and Google positions it as the fast, everyday tier. Carbon and Argon are the frontier models, built for long-horizon reasoning, complex multi-step coding, and scientific research. Hydrogen is built for execution: writing the change, running the test, reading the failure, and trying again, as Google puts it. It accepts text, images, audio, and video, and it works across a 1 million-token context window, which is the same size as the rest of the Gemini 4 family. It has four thinking levels — low, medium, high, and a new extra-high level called xhigh — so a developer or an app can trade reasoning depth for speed and cost on each task. The high level is the default. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at that default setting, more than triples the older model's score on Terminal-Bench 4.1, a test of real coding-agent work, from 18.2 percent to 57 percent, and cuts fabricated claims by more than five times. On the Artificial Analysis Intelligence Index, a broad composite benchmark, Google reports Hydrogen scoring 51 at its default high thinking level and 54 at xhigh, one point above Gemini 4 Argon at 50. Google also cites a fifth-place ranking on Agent Arena for multi-step work, a leading score on the HLE-Diamond benchmark at 58.8 percent, and first place on the Vals Tax Agent Bench at 52.3 percent. All of those numbers are vendor-reported. None has been independently audited, and the model card itself supplies a useful qualification: on the cheating-adjusted Terminal-Bench 4.1 table, Claude Sonnet 5.5 scores 63.9 percent, above Hydrogen's 57.0 percent. The honest reading is that Hydrogen is a large step up from the model it replaces, and still not the strongest model in its class on every test.
What changes in Google Search
For most people, the first place Hydrogen shows up is Search, and that is where Google is making its boldest claim. Stein's post says Hydrogen is available globally in Google Search for all users, and that it is now the new default model in AI Mode and is used for the hardest questions in AI Overviews. In practice, Google says Search should now handle questions with several parts, highly specific requests, and hard-to-read typos, answering each part of a query instead of only the part it understood best. The model also powers Search's query fan-out technique, which splits one prompt into several subqueries run at the same time, and Google says the result should be more helpful links, more visual responses — images, structured responses, comparison tables, and custom tools — and better answers across the languages supported by AI Mode. The change is easy to miss, because nothing in the Search interface announces it. There is no new button and no model picker in Search. The upgrade arrives silently, which is the point: Google is moving the highest-volume AI workload on earth, the questions billions of people type into a search box, onto a Gemini 4 model. In the Gemini app, the change is slightly more visible. Hydrogen appears in the model picker for paying subscribers, while free users receive it through automatic model selection, when Google decides a prompt suits it. The distinction matters, because it means most free users will use Hydrogen without ever choosing it, and the quality of their answers will depend on Google's routing decisions as much as on the model itself.
Pricing, and who gets it
Hydrogen is priced flat across its entire 1 million-token context window at 1.50 dollars per million input tokens and 7.50 dollars per million output tokens, with a rate-limited free tier in the Gemini API and AI Studio, and implicit caching at 0.075 dollars per million cached tokens. That flat price is the part of the announcement that developers will argue about most. It is double what Gemini 3.8 Flash costs today, and equal to what 3.8 Flash will cost from January 1, 2027, when its own price rises. It is also, by Google's framing, the cheapest tier in the Gemini 4 line, and roughly 25 percent below the introductory rate Google is still charging for Gemini 4 Argon. The comparison that matters for builders is not the headline rate but the total cost of a task, because Hydrogen's four thinking levels change how many tokens a job consumes. A model that finishes a coding task 2.5 times faster may also burn fewer output tokens, which is where most of the cost sits. Google has not published a worked example of that arithmetic, so developers will need to test it on their own workloads. Access is broad: all tiers of Google AI Studio and the Gemini API, Gemini for business, the Gemini app, Google Search, and Google Antigravity for paid Google AI Pro and Ultra subscribers. One constraint worth knowing is in the documentation rather than the announcement. Hydrogen is a short-term generally available release, subject to a minimum 45-day retirement notice, and the Cloud model page lists explicit context caching, RAG Engine support, tuning, and the Gemini Live API as not supported. The knowledge cutoff is March 2026, so the model does not know about events after that date unless a tool or retrieval step supplies them.
Orchestrator: Carbon plans, Hydrogen executes
Alongside Hydrogen, Google introduced Orchestrator, a model option in Antigravity that pairs Hydrogen with Gemini 4 Carbon. The idea is a division of labour: Carbon, the frontier model, plans the work, and Hydrogen, the fast model, executes it. Google says Orchestrator reached near-Carbon task success rates on Terminal-Bench 4.1, which, if it holds up in real use, would let developers get frontier-level results while paying mostly for the cheaper model's tokens. Orchestrator is rolling out first to Google AI Ultra subscribers, with enterprise access through Gemini for business. It is the clearest sign of where Google thinks the model market is heading: not one model for everything, but a small team of models, with a cheap, fast worker doing most of the steps and an expensive thinker stepping in only where it is needed. The same pattern is visible across the industry, but Google is the first to ship it as a named product inside its own coding tool. Whether the routing works well enough to trust is an open question, and the only way to answer it is to run real tasks through it and check the results, because a planner that hands the wrong step to the wrong model will produce confident-looking work that is wrong in ways that are hard to spot.
Checklist: should you switch to Hydrogen?
- If you use Google Search or the Gemini app, you do not need to do anything: Hydrogen is already rolling out to you, and free users get it through automatic model selection.
- If you build with the Gemini API or AI Studio, test Hydrogen against Gemini 3.8 Flash on your own tasks before switching, and measure total cost per finished task, not just the price per token.
- If you run coding agents, try the four thinking levels on the same task: low for simple edits, high for the default, and xhigh where accuracy matters most, and watch how the token bill changes.
- If you need retrieval, tuning, or the Live API, check the model documentation first: Hydrogen does not support RAG Engine, tuning, or the Live API, and only implicit caching is available.
- If you are a Google AI Ultra subscriber, try Orchestrator in Antigravity, but verify its output the same way you would verify any agent's work, because the handoff between Carbon and Hydrogen is new.
- If you depend on a model staying available for months, note the short-term availability status and the 45-day minimum retirement notice, and keep a fallback model in your configuration.
- If you care about benchmark claims, treat every number in this article as Google's claim, not an independent result, and wait for third-party testing before quoting them as fact.
What to watch next
Three things will decide whether Hydrogen is a genuine step forward or just a faster version of the same trade-offs. The first is independent testing. Google's numbers are strong, but they are Google's numbers, measured on Google's harness, on the day of release. The Artificial Analysis index, Agent Arena, and Terminal-Bench are all public benchmarks, so independent runs should appear within days, and the gap between Google's claims and those runs will tell you how much to trust the rest of the announcement. The second is the Search experience itself. Hydrogen is now the default in AI Mode and the model behind the hardest AI Overview questions, so the quality of everyday search answers is the largest real-world test of the model, and it is running in front of billions of users. Watch for whether answers to multi-part questions actually improve, and whether the extra visual responses help or clutter. The third is price. Hydrogen costs twice what 3.8 Flash costs today, and the Gemini 4 family has a pattern: Argon's introductory rate roughly doubled after its introductory period, and Carbon's price is scheduled to double on December 1, from 2 dollars to 4 dollars per million input tokens. Hydrogen's flat rate looks cheap now. The question is how long it stays flat, and whether the 45-day retirement notice is the real deadline to plan around. Google has now shipped three Gemini 4 tiers in five weeks. The pace is the story as much as any single model.
Bottom line
Google launched Gemini 4 Hydrogen on November 12, 2026, as the fast, everyday member of the Gemini 4 family, replacing Gemini 3.8 Flash across the Gemini app, Google Search, the Gemini API, AI Studio, Gemini for business, and Antigravity. Google says it finishes tasks 2.5 times faster than the model it replaces, scores 57 percent on Terminal-Bench 4.1 against 18.2 percent, and reaches 51 on the Artificial Analysis Intelligence Index at its default setting, one point above the frontier model Argon. Its biggest change is invisible: it is now the default model in AI Mode and the model behind the hardest questions in AI Overviews, which puts a Gemini 4 model in front of every Google Search user. Pricing is flat at 1.50 dollars per million input tokens and 7.50 dollars per million output tokens, with a rate-limited free tier, but the model is a short-term release with a 45-day minimum retirement notice and no support for RAG Engine, tuning, or the Live API. Every benchmark in this article is Google's own claim, and independent testing will be the real verdict. If you use Search or the Gemini app, you are already using Hydrogen. If you build with it, test it on your own tasks and measure the cost per finished job before you commit.
Reference sources
- Google — Introducing Gemini 4 Hydrogen for Agentic Workflows, 12 November 2026
- TechCrunch — Google's latest Search upgrade promises better answers to your most complicated questions, 12 November 2026
- Unite.AI — Google Launches Gemini 4 Hydrogen for Everyday Agentic Workflows, 12 November 2026
- MakeUseOf — Google's new Gemini 4 Hydrogen is rolling out now, and free users get it too, 12 November 2026
- Silicon Report — Google launches Gemini 4 Hydrogen, replacing Gemini 3.8 Flash, 12 November 2026
- LLM Report — Google releases Gemini 4 Hydrogen, a fast model that replaces 3.8 Flash and upgrades Search, 12 November 2026
Leave a public comment