Gemini 4 Hydrogen Launch: $1.50 Model Powers Search Now

Google launched Gemini 4 Hydrogen on November 12, 2026, and it is the most useful Gemini 4 model for most people — not because it is the smartest, but because it is the fast, cheap one that shows up where you already work. It costs $1.50 per million input tokens and $7.50 per million output tokens, flat across its entire 1-million-token context window. Google says it runs 2.5 times faster than Gemini 3.8 Flash, the model it replaces. And unlike its bigger siblings Carbon and Argon, Hydrogen is rolling out to everyone — free Gemini app users and all Google Search users included.

If you build with AI or pay API bills, this launch changes your cost math this week. Below: the full family price table, the worked cost of a real coding session on each model, which model fits which job, and the one catch in the fine print — a 45-day retirement notice.

What Google actually announced

Koray Kavukcuoglu, senior vice president at Google DeepMind and Google’s chief AI architect, announced Hydrogen on Google’s blog on November 12 under the title “Introducing Gemini 4 Hydrogen for Agentic Workflows.” The positioning is deliberate: Hydrogen is the execution engine, not the strategist. Google describes it as built for “rapid iteration” — writing code, running tests, verifying changes, doing research, and carrying multi-step tasks to completion.

That framing matters because Google now has three Gemini 4 models with three different jobs. Argon is the general-purpose flagship. Carbon is the deep-reasoning model for the hardest jobs. Hydrogen is the fast one that does the everyday work cheaply — the drafting, debugging, summarizing, and iterating that make up most of a freelancer’s or small team’s AI use.

The launch also landed in Google Search the same day. Robby Stein, Google’s vice president of product for Search, called it “the biggest model upgrade to Search yet.” Hydrogen is now available globally in Google Search for all users, not only subscribers. That is a real change from thirty days earlier, when Google put Argon in Search only for paying subscribers. For the first time, a Gemini 4 model is answering questions for free users. (For the full Search-side picture, see PPC Land’s breakdown of the rollout.)

The Gemini 4 family, side by side

Here is the whole family as it stands on November 13, 2026, with prices taken from Google’s own announcement and pricing pages:

Two things jump out of that table. First, Hydrogen is the only Gemini 4 model with a flat price — no introductory rate that doubles on December 1. Carbon and Argon both cost $2/$10 through November 30, 2026, then jump to $4/$20 on December 1 — the full doubling schedule is covered in our Gemini 4 Argon pricing guide. Hydrogen’s $1.50/$7.50 is the price, full stop, at least as Google has published it so far.

Second, Hydrogen is the only Gemini 4 model free users can reach. Carbon went generally available on November 10, but only to paying API developers and Google AI Ultra subscribers. Argon has been generally available since October 13, again on the paid tier. Hydrogen arrives on every tier of AI Studio and the Gemini API, in the Gemini app, in Gemini for business, and in Search — and free users get it through automatic model selection, when Google’s system decides it suits the prompt.

One honest caveat on the table: the Carbon and Argon price-doubling dates are confirmed on Google’s pricing page, but Hydrogen’s long-term pricing has no published end date or guarantee. Treat $1.50/$7.50 as the launch price, not a forever price.

What the benchmarks actually say — and what they don’t

Google’s announcement leans on three numbers, all Google’s own measurements — not independent results:

• Artificial Analysis Intelligence Index: Hydrogen scores 51 at its default high thinking level and 54 at its highest xhigh level, against 50 for Argon. Yes — the cheap fast model edges the flagship on this particular index.
• Terminal-Bench 4.1: Hydrogen scores 57%, against 18.2% for Gemini 3.8 Flash. That is the agentic-coding benchmark, and it is the number that matters most for developers.
• Speed: 2.5 times faster than Gemini 3.8 Flash at default settings.

The Artificial Analysis result is the headline grabber, and it deserves a second look. Argon is the bigger, more expensive model. Hydrogen beating it on one index does not mean Hydrogen is smarter — it means the index rewards the kind of fast, competent execution Hydrogen is tuned for. Argon still does the long, hard reasoning jobs Hydrogen is not built for. Nobody outside Google has independently replicated any of these numbers yet. If you are choosing a model for production work this week, treat the benchmarks as a directional signal and test Hydrogen on your own workload before switching anything that matters.

Google also gives Hydrogen four thinking levels — low, medium, high (the default), and a new xhigh — so you can trade quality against speed and cost: a quick draft runs on low, a tricky debugging loop on xhigh.

What Hydrogen really costs: the worked math

Per-token prices are abstract until you run them through a real workload. Here is a realistic one: a coding session where you paste in a 80,000-token codebase context and the model writes back 20,000 tokens of code, tests, and explanation.

• Hydrogen: 80,000 input tokens × $1.50 per million = $0.12. 20,000 output tokens × $7.50 per million = $0.15. Total: $0.27 per session.
• Carbon or Argon, November pricing ($2/$10): $0.16 + $0.20 = $0.36 per session.
• Carbon or Argon, December pricing ($4/$20): $0.32 + $0.40 = $0.72 per session.

Hydrogen is 25% cheaper than the flagship models this month and 62% cheaper from December 1. Scale it up: a developer running 1,000 such sessions a month pays roughly $270 on Hydrogen, $360 on Carbon or Argon in November, and $720 on either flagship from December. That is a $450-a-month difference in December, for work Hydrogen is specifically designed to do. Flip it around: $10 buys about 37 of those sessions on Hydrogen, 27 on a flagship in November, and only about 14 from December.

The flat-rate context window is the other half of the savings. Hydrogen charges the same $1.50/$7.50 across the whole 1-million-token window — no long-context surcharge past 128,000 or 200,000 tokens, the way some models price large prompts. A 900,000-token repository dump costs the same per token as a 10,000-token question, and agentic workflows are exactly the ones that stuff huge contexts into the model.

One more comparison worth knowing: according to pricing trackers that read Google’s own price lists, Hydrogen’s $1.50/$7.50 is roughly twice what Gemini 3.8 Flash costs today — and the same as what Flash is scheduled to cost from January 1, 2027. In other words, Google is not giving you a cheaper Flash. It is giving you a much faster, more capable model at the price Flash is about to reach anyway.

Which Gemini 4 model for which job

The family only makes sense as a routing decision. Here is how to think about it:

Use Hydrogen when: you are writing everyday code, running tests, debugging, doing research, summarizing documents, or running an agent through a multi-step task. This is the volume work — the dozens of small calls a day that add up on the bill — and the right default for anything user-facing where latency matters.

Use Argon when: you need the general-purpose flagship — the strongest all-rounder for complex coding, knowledge work, math, and science, where you want the best single answer and you are willing to pay for it. Argon is the default for most software engineering and knowledge work, in Google’s own positioning.

Use Carbon when: the job is genuinely hard — long multi-step coding and debugging, open-ended scientific research, or security work where Carbon’s extra thinking pays off. Carbon deliberately spends more tokens than Argon exploring, reasoning, and running verification loops, and it can be slow or time out on the hardest jobs. Budget Carbon at the December $4/$20 rate and treat the November $2/$10 as a bonus.

Use the thinking levels: low for drafts and summaries, medium for routine code, high (the default) for real work, xhigh for the problem that has already beaten two other attempts. The level is a cost lever as much as a quality lever.

The practical move is a routing rule, not a loyalty choice: Hydrogen by default, Argon when the answer matters, Carbon when nothing else has cracked it. Most teams will find 80% of their calls land on Hydrogen — and their bill will show it.

Where you can use Hydrogen today

The rollout is broad, but the access rules differ by surface. Here is the honest map:

• Google Search (free, everyone): Hydrogen is live globally for all users. You do not pick the model — Google routes your query to it. This is the first Gemini 4 model free users have ever touched.
• Gemini app (free and paid): Hydrogen arrives through automatic model selection. Free users get it when Google’s system decides it fits the prompt.
• Gemini API and AI Studio (all tiers, including free tier): Developers can call the model directly. This is where the $1.50/$7.50 pricing applies.
• Gemini for business: Available to enterprise customers through the business tier.
• Antigravity (Google AI Pro and Ultra subscribers): Hydrogen is available in Google’s agentic coding environment — and this is where the most interesting feature lives.

Orchestrator: Carbon plans, Hydrogen executes

Inside Antigravity, Google introduced a feature called Orchestrator. It hands work between Carbon and Hydrogen inside a single session: Carbon does the deep planning and reasoning, then Hydrogen does the fast execution — writing the code, running the tests, verifying the changes.

That is the multi-model workflow in miniature, and it is the pattern Google is betting on. The expensive model thinks; the cheap model works. If you have ever watched an agent burn a flagship model’s tokens on routine file edits and test runs, you know why this matters. Orchestrator automates the routing decision described above, inside the tool where developers actually live. It is available to paid Pro and Ultra subscribers in Antigravity — not a free-tier feature. But the pattern is portable: you can build the same routing yourself in any API workflow, and the savings are the same.

The catch: a 45-day retirement notice

Google Cloud’s model documentation page lists Hydrogen as a short-term generally available (GA) release, subject to a minimum 45-day retirement notice. In plain English: Google can retire this model with as little as 45 days’ warning.

That is not a dealbreaker — 45 days is enough to migrate a well-built integration — but it is a real constraint, and it is the number a buyer should place next to the price. Do not hardcode the model name into anything you cannot change quickly: keep it in a config value, not buried in code. Budget for the possibility that Hydrogen’s price moves, and do not build a business that only works if Hydrogen exists at this price. If your margins depend on $1.50/$7.50, have a fallback model priced and tested — Argon at $4/$20 is the obvious one.

The 45-day notice is also a signal about Google’s release cadence: Gemini 4 is moving fast — Argon in October, Carbon in November, Hydrogen in November — and Google is clearly willing to ship, learn, and replace. That speed is good for users. It is also a reason to build loosely coupled integrations.

What Hydrogen cannot do (yet)

The documentation adds four limits the announcement does not emphasize:

• No image or audio generation. Hydrogen is a text and multimodal-understanding model. If your workflow needs the model to generate images or audio, that is a different tool.
• No Live API. There is no real-time, low-latency voice or video conversation mode for Hydrogen.
• No explicit context caching. Hydrogen supports implicit caching only, with a 4,096-token minimum. Explicit cache control is not supported, though Google says TTL-based cache control arrives for Hydrogen and later models starting January 12, 2027. If your cost model depends on aggressive caching, check the docs before you commit.
• No RAG Engine or tuning support. You cannot fine-tune Hydrogen or use Google’s RAG Engine with it, according to the model documentation.

The knowledge cutoff is March 2026, and the unified context limit is 1,048,576 tokens including both input and output — the “1 million” in the marketing is a rounded figure.

None of these limits is a flaw. They are the shape of a fast execution model. But they are the difference between a smooth adoption and a surprised one, and they are exactly the details that get lost in launch-day coverage.

What early testers saw before the launch

Hydrogen did not appear out of nowhere. For about two weeks before the announcement, community testers reported meeting a fast new model in blind LMArena matchups under the label gemini-3.8-flash, and early demos showed it producing detailed 3D scenes in minutes. One Reddit user reported a side-by-side test in which Hydrogen finished a motion-graphics prompt in about 2 minutes while Gemini 3.8 Flash took about 11 minutes. Another early tester described the jump as comparable to a full model-generation leap, not an incremental update.

Treat all of that as community signal, not fact — none of it had a benchmark, a model card, or a Google source behind it before launch day. What the chatter does tell you is that the speed story is consistent: Google’s official 2.5x figure, the testers’ anecdote, and the Terminal-Bench gap all point the same direction. The lesson for anyone following AI releases: the rumor cycle spotted this model weeks early through billing catalogs and arena matchups, so by launch day the price was already known. The same week also brought Qwen4-Max’s preview launch — the model race is moving too fast to follow from announcements alone.

What this means for freelancers and small teams

Strip away the launch language and Hydrogen is a cost-structure story. A freelancer running an agentic coding loop — plan, write, test, fix, repeat — was paying flagship prices for every step, including the cheap steps. Hydrogen lets you pay flagship prices only for the thinking and fast-model prices for the doing. For a solo developer, that can be the difference between a $300 monthly API bill and a $120 one.

For small agencies and content teams, the Search rollout matters just as much as the API price. Hydrogen now powers the answers Google shows everyone, which raises the quality bar for AI-written answers in Search. If your content strategy depends on being the source Search cites, the way to stay citable is the same as ever: direct answers, verifiable facts, named sources. A better model in Search makes that more important, not less.

What to do this week

1. Try it free first. Open the Gemini app or ask a question in Google Search — Hydrogen is already there. Run a real task through it: a debugging session, a document summary, a research pass. Compare it against what you used last week.
2. Check your API routing. If you call the Gemini API, look at which model your code defaults to. If a flagship model is doing routine work, route that work to Hydrogen and measure the bill for a week.
3. Price the December jump. If you use Carbon or Argon, your per-token price doubles on December 1. Run the worked math above against your actual usage and decide now whether Hydrogen covers the jobs you are paying flagship prices for. Our AI price deadlines roundup tracks every November model-price change in one place.
4. Put the model ID in config. If you integrate Hydrogen, keep the model name in a single config value. The 45-day retirement notice is only a problem if migration is hard.
5. Watch for independent benchmarks. Google’s numbers are a starting point. Real evaluations from independent testers will land over the next two weeks — check them before moving anything production-critical.

The bottom line

Gemini 4 Hydrogen is not the most powerful model Google has ever shipped, and Google is not pretending it is. It is the fast, cheap, everywhere model — $1.50 in, $7.50 out, 2.5 times faster than the Flash it replaces, available to free users for the first time in the Gemini 4 family, and already answering questions in Google Search for everyone. The price is honest, the speed claim is plausible and consistent with what testers saw before launch, and the family math makes Hydrogen the default for most everyday AI work.

The two things to remember are the ones the launch post does not headline: Carbon and Argon double in price on December 1, which makes Hydrogen’s flat rate the cheapest Gemini 4 option by a wide margin from next month — and Hydrogen is a short-term GA release with a 45-day retirement notice, so build your integration to switch models easily. Get those two right and Hydrogen is a genuine saving.

Frequently Asked Questions

What is Gemini 4 Hydrogen?

Gemini 4 Hydrogen is Google’s fast, cost-efficient Gemini 4 model, launched November 12, 2026. It is built for everyday coding and agentic workflows — writing code, running tests, verifying changes, and carrying multi-step tasks to completion. It costs $1.50 per million input tokens and $7.50 per million output tokens, flat across its 1-million-token context window, and it replaces Gemini 3.8 Flash.

How much does Gemini 4 Hydrogen cost?

Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens through the Gemini API. A typical coding session with 80,000 input tokens and 20,000 output tokens costs about $0.27. The price is flat across the entire 1-million-token context window, with no long-context surcharge. By comparison, Carbon and Argon cost $2/$10 through November 30, 2026, then rise to $4/$20 on December 1.

Is Gemini 4 Hydrogen free to use?

Yes, in the consumer surfaces. Hydrogen is rolling out in the Gemini app and Google Search for all users, including free users, through automatic model selection. API access is paid at the $1.50/$7.50 rate, and Antigravity’s Orchestrator feature needs a paid Google AI Pro or Ultra subscription.

How is Hydrogen different from Carbon and Argon?

Hydrogen is the fast execution model; Argon is the general-purpose flagship; Carbon is the deep-reasoning model for the hardest jobs. Hydrogen is cheaper ($1.50/$7.50 vs $2/$10 rising to $4/$20), faster (2.5x the speed of Gemini 3.8 Flash), and the only one available to free users. Carbon and Argon do the heavy thinking; Hydrogen does the everyday work. Google’s Orchestrator feature in Antigravity splits work exactly this way — Carbon plans, Hydrogen executes.

What are the limits of Gemini 4 Hydrogen?

Hydrogen has no image or audio generation, no Live API, no explicit context caching, and no RAG Engine or tuning support. Its knowledge cutoff is March 2026. It is also a short-term GA release Google can retire with 45 days’ notice — keep the model ID in config and have a fallback tested.

Should I switch from Gemini 3.8 Flash to Hydrogen?

Yes, if you use Flash for coding or agentic work. Google positions Hydrogen as Flash’s replacement, it is 2.5 times faster at default settings, and it scores 57% on Terminal-Bench 4.1 against Flash’s 18.2% (both figures are Google-reported). The price is roughly twice Flash’s current rate but matches what Flash is scheduled to cost from January 2027, so you are paying the incoming Flash price for a much faster model. Test it on your own workload first, and watch for independent benchmarks over the next two weeks.

[Sources: Google’s announcement, “Introducing Gemini 4 Hydrogen for Agentic Workflows” (blog.google, November 12, 2026); Google Search announcement by Robby Stein (blog.google, November 12, 2026); Google Cloud model documentation for Gemini 4 Hydrogen; Google’s Gemini API pricing page. All benchmark figures are Google-reported and not independently verified.]