Gemini 4 Hydrogen: Google’s Fast Execution Model Explained

Google has given its Gemini 4 family a third member — and this one is built to work fast, not think longest. Gemini 4 Hydrogen, launched on 12 November 2026, is the execution model of the family: the one that writes the code, runs the tests and finishes the everyday jobs, while Carbon and Argon keep the deep, long-horizon reasoning. It is already rolling out in Google Search, the Gemini app, AI Studio and the API, and it costs $1.50 per million input tokens and $7.50 per million output tokens, flat across its full one-million-token window.

Quick answer: Gemini 4 Hydrogen is Google's fast Gemini 4 model for coding. Launched 12 November 2026, it costs $1.50/$7.50 per 1M tokens and scores 51 on the AA Index.

The launch matters for a simple reason. Until now, the Gemini 4 family had a gap: Carbon was the heavy reasoner and Argon the everyday all-rounder, but neither was built to be the cheap, fast workhorse you point at a hundred small jobs a day. Hydrogen is that workhorse — and Google is putting it in front of everyone at once, including free Search users. This explainer covers what Hydrogen is, what it costs per finished task rather than per token, how fast Google says it is, and when you should — and should not — switch to it.

What is Gemini 4 Hydrogen?

Gemini 4 Hydrogen is Google’s fast, mid-priced member of the Gemini 4 family, released on 12 November 2026. Google describes it as built for “rapid execution across diverse agentic workflows”: writing code, running tests, verifying changes and carrying out multi-step tasks. It is adapted from Gemini 4 Carbon, according to the model card, and its knowledge cutoff is March 2026. It is a proprietary model — there are no open weights.

The family split is now explicit. Carbon is the deep reasoner for your hardest problems, Argon sits in the middle, and Hydrogen is the executor. Google’s own framing is that teams still need frontier models like Carbon and Argon for architectural judgement and untangling subtle bugs, but they also need a fast, disciplined model for the everyday work around them. Hydrogen is meant to stand on its own for that work — or to pair with Carbon and Argon as the high-efficiency execution engine in a multi-model workflow.

Model Role in the family Artificial Analysis Index API price per 1M tokens
Gemini 4 Hydrogen Fast execution: coding, tests, multi-step agentic work 51 at default high; 54 at xhigh $1.50 in / $7.50 out, flat
Gemini 4 Carbon Deep, long-horizon reasoning 59 $2 / $10 introductory, rising to $4 / $20 from 1 December 2026
Gemini 4 Argon Everyday reasoning and standard engineering 50 See our comparison below

The index scores come from Artificial Analysis, an independent leaderboard that Google itself cites. Hydrogen beating Argon by a point at its default setting is the headline number, but the honest reading is narrower than the launch copy suggests: Hydrogen is not the strongest Gemini 4 model. Carbon still holds that title. Hydrogen’s claim is speed and price per finished task, not raw capability — and that is the comparison that matters for most users. For the full Carbon picture, read our Gemini 4 Carbon explainer, and for how Argon and Carbon differ, see our Argon vs Carbon comparison.

How much does Gemini 4 Hydrogen cost?

Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, flat across its entire one-million-token context window. There is no introductory rate and no dated price step — unlike Carbon and Argon, whose introductory rates double on 1 December 2026. Batch and Flex requests cost half the standard rate, according to Google’s pricing page. There is also a rate-limited free tier on the Gemini API and in AI Studio, and Hydrogen is free in Google Search and, by automatic selection, in the free Gemini app.

Token prices alone do not tell you what a job costs, because models use different numbers of tokens to finish the same task. So here is a worked illustration using published prices only, with the token counts stated as an assumption rather than a measurement. Take a coding job that needs 100,000 input tokens and 20,000 output tokens:

Model Input (100,000 tokens) Output (20,000 tokens) Total for the job
Gemini 4 Hydrogen $0.15 $0.15 $0.30
Gemini 4 Carbon (introductory rate) $0.20 $0.20 $0.40
Gemini 4 Carbon (from 1 December 2026) $0.40 $0.40 $0.80

On identical token counts, Hydrogen is 25% cheaper than Carbon today and 62% cheaper once Carbon’s introductory rate ends. The catch is the assumption: if Carbon finishes the same job with far fewer tokens, or Hydrogen needs several retries, the ranking can flip. Google’s own partner testing says Hydrogen ran 2 to 2.8 times faster while using 12% to 23% fewer tokens than Gemini 3.8 Flash — but that comparison is against Flash, not against Carbon, so treat the table as a price illustration, not a measured cost test. The practical rule is to compare cost per finished task, not cost per token.

How fast is Hydrogen, really?

Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default high thinking setting, and more than triples Flash’s Terminal-Bench 4.1 score — 57% against 18.2%. It also says Hydrogen cuts fabricated claims by more than five times compared with Flash, and adapts cleanly when a tool call fails or a search returns empty. Every one of those figures is Google’s own, from its launch announcement and its partner testing. No independent end-to-end benchmark of Hydrogen against rival models had been published as of 13 November 2026, so treat the speed figures as vendor claims, not settled measurements.

The independent anchors that do exist are modest but real. Artificial Analysis scores Hydrogen at 51 on its Intelligence Index at the default high setting — one point above Argon — and 54 at the highest xhigh setting. On Agent Arena, a live benchmark, Hydrogen ranks fifth. Google also reports that Hydrogen leads its class on HLE-Diamond at 58.8%, ranks first overall on the Vals Tax Agent Bench at 52.3%, ties for first on the Vals Legal Research Bench at 55.8%, leads its class on AutomationBench-AA, scores 53.4% on the Multimodal Knowledge Agent benchmark and 71.6% on Agents’ Last Exam v1.1. Those are Google-reported results on third-party benchmarks, which is a stronger signal than a vendor-only chart, but they are still numbers Google chose to publish.

One number deserves a pause. Google’s launch chart has Hydrogen beating Claude Sonnet 5.5 on seven of ten tests — but Sonnet 5.5 leads the other three, and the chart is Google’s own selection. The honest summary is that Hydrogen is competitive in its class and clearly faster than the model it replaces, not that it has beaten the frontier.

What do the four thinking levels do?

Hydrogen offers four thinking levels — low, medium, high and a new xhigh tier — with high as the default. Google says they scale in a smooth, predictable progression, so you only wait and pay for the reasoning depth a task actually needs.

Thinking level Use it for What you get
Low Quick lookups, short replies, simple rewrites Fastest, cheapest answers
Medium Everyday drafting, summaries, comparisons A balance of speed and care
High (default) Coding, debugging, multi-step agentic work The setting behind the 51 index score and the 2.5x speed claim
Xhigh The hardest jobs you would still give a fast model Index score rises to 54, at higher cost and latency

The level matters more than it looks. At low, Hydrogen is a quick-answer model. At xhigh, it is a different machine on the same price list — and the 54 index score is the number Google leads with when it wants Hydrogen compared with stronger models. If you try Hydrogen and find it shallow, check the thinking level before you blame the model: the default high setting is where Google’s headline figures were measured.

What is Orchestrator in Antigravity?

Alongside Hydrogen, Google launched Orchestrator in Antigravity, its agent-focused development platform, for paid Google AI Pro and Ultra subscribers. Orchestrator dynamically switches between models like Carbon and Hydrogen within the same session: Carbon plans, Hydrogen executes. Google reports that on Terminal-Bench 4.0, Orchestrator reached near-Carbon task success rates while consuming 24% less quota and completing tasks 19% faster on average. For workflows with parallel agents, Carbon can act as coordinator while Hydrogen powers worker and verifier agents, building and testing changes simultaneously across isolated workspaces.

This is the most interesting part of the launch, because it makes the family split operational. Instead of choosing one model for a whole job, Orchestrator routes each step to the model that fits it — planning to the expensive reasoner, execution to the cheap fast one. The 24% quota saving and 19% speed gain are Google’s figures from its Antigravity announcement, and Orchestrator is Ultra-only at launch, so most developers will not be able to try it yet. But the pattern it represents — a planner model plus a worker model, routed automatically — is likely to become the default way agentic coding tools are built.

Where can you use Hydrogen today?

Hydrogen is rolling out globally across Google’s developer, enterprise and consumer surfaces. The rollout started on 12 November 2026, but not every account sees it in every menu yet — the Gemini app, in particular, is still rolling out, and free app users only get Hydrogen when Google’s automatic model selection picks it.

Surface Who gets it Status
Gemini API / AI Studio All tiers, including a rate-limited free tier Available now; model ID gemini-4-hydrogen
Google Search All users, globally Rolling out now
Gemini app Paid plans in the model menu; free users via auto-selection only Rolling out now
Gemini for business Business subscribers, via API or model menu Available now
Google Antigravity Paid Google AI Pro and Ultra subscribers Available now; Orchestrator is Ultra-only
Gemini in Sheets / Android Studio Eligible business subscriptions; developers Rolling out now

For consumers, the change is mostly invisible: Hydrogen powers better Search responses — Google says it understands multi-part queries and typos better, returns more visual responses, and fans out to a wider range of links — and a more reliable Gemini app that follows instructions more closely across multi-turn conversations. For developers, the practical entry point is the API in AI Studio, where the model ID is gemini-4-hydrogen and the one-million-token context window is shared across input, thinking and output. Hydrogen is a text model, so for AI video the relevant Google launch is different: see our guide to Gemini Omni 1.2 Flash. And if your job is scoring a thousand yes-or-no calls rather than writing code, our guide to the Gemini Decisions API covers the tool Google built for exactly that.

Should you switch from Gemini 3.8 Flash?

Google calls Hydrogen the recommended upgrade from Gemini 3.8 Flash and earlier models, and the Flash name is gone from the new family. Three questions decide whether to switch today:

  1. Are you on 3.8 Flash now? If so, switch. Hydrogen is faster on every measure Google publishes, cuts fabricated claims by more than five times, and is the model Google will keep developing. Staying on Flash means staying on a superseded model.
  2. Do your jobs need deep reasoning or fast execution? If the job is planning an architecture, untangling a subtle bug or reasoning over a long horizon, Carbon is still the right model — Hydrogen is the executor, not the thinker. If the job is writing code, running tests and verifying changes, Hydrogen is the cheaper, faster pick.
  3. Are you paying Carbon prices for routine work? If you have been routing everyday jobs to Carbon, Hydrogen at $1.50/$7.50 is the obvious downgrade in price without a downgrade in the work that matters. The worked example above shows the gap widening after 1 December, when Carbon’s introductory rate ends.

One warning before you trust any listing you find. There is no Gemini 4 Pro. No such model identifier, price or model card exists in Google’s documentation — and from 1 December 2026, API requests to the old gemini-2.5-pro identifier will be served by Gemini 3.8 Flash instead. If a tool, tutorial or price list promises you “Gemini 4 Pro”, it is describing something that does not exist. Check the model ID: gemini-4-hydrogen.

Our take

The real story of this launch is not a benchmark number. It is that Google has stopped selling one model for everything and started selling a team: Carbon plans, Hydrogen executes, Orchestrator routes. That split is honest about how agentic coding actually works — most steps in a coding job are routine execution, and paying frontier prices for routine execution is waste. Hydrogen’s price, $1.50 in and $7.50 out with no introductory rate to expire, is the clearest signal of intent in the announcement: Google wants this model to be the default, not the discount.

Judge Hydrogen on cost per finished task, not cost per token — and not on Google’s chart alone. The 2.5x speed claim, the 57% Terminal-Bench score and the safety figures are all Google’s own. What is independently verifiable today is the price, the availability and the 51 on the Artificial Analysis Index. Everything else is a claim to re-test once independent benchmarks land.

Two things to watch next. First, whether independent benchmarks confirm the speed story once Hydrogen has been in developers’ hands for a few weeks. Second, whether the planner-plus-worker pattern spreads beyond Antigravity — if Orchestrator’s routing works, every coding agent will want the same trick, and the question will stop being “which model is best” and become “which model for which step”. That is a better question, and Hydrogen is Google’s answer to the second half of it.

Desk-researched, not hands-on tested: this article is based on Google’s announcement and independent coverage read on 13 November 2026. No independent benchmark of Hydrogen against rival models had been published at the time of writing, and all Google-reported figures are labelled as such.

Frequently Asked Questions

Is Gemini 4 Hydrogen free?

Partly. Hydrogen is free in Google Search, free in AI Studio and the Gemini API on a rate-limited free tier, and free in the Gemini app when Google’s automatic model selection picks it. Paid API use costs $1.50 per million input tokens and $7.50 per million output tokens, and paid Gemini app plans can select it directly from the model menu.

Is Hydrogen better than Carbon?

No — they do different jobs. Carbon is the stronger reasoner, scoring 59 on the Artificial Analysis Intelligence Index, and it is the right model for deep, long-horizon work. Hydrogen is the faster executor, scoring 51 at its default high setting and 54 at xhigh, and it is the right model for everyday coding and agentic work. Use Carbon to plan and Hydrogen to build.

What is the Gemini 4 Hydrogen model ID?

The model ID is gemini-4-hydrogen. Use it in the Gemini API and AI Studio. Its knowledge cutoff is March 2026, and it shares a single 1,048,576-token window across input, thinking and output.

Does Hydrogen replace Gemini 3.8 Flash?

Yes. Google calls Hydrogen the recommended upgrade from Gemini 3.8 Flash and earlier models, and the Flash name is dropped from the new family. Hydrogen is faster on every measure Google publishes and cuts fabricated claims by more than five times compared with Flash.

Is there a Gemini 4 Pro?

No. There is no Gemini 4 Pro model identifier, price or model card in Google’s documentation. From 1 December 2026, API requests to the old gemini-2.5-pro identifier will be served by Gemini 3.8 Flash instead. The Gemini 4 family is Carbon, Argon and Hydrogen.

Sources and methodology

Sources: Google — Introducing Gemini 4 Hydrogen for Agentic Workflows (12 November 2026); Unite.AI — Google Launches Gemini 4 Hydrogen (12 November 2026); Kingy — What You Actually Get Today (12–13 November 2026); Artificial Analysis — Intelligence Index; OrcaRouter — Gemini 4 Pro Does Not Exist (13 November 2026); Magica — Google Launches Gemini 4 Hydrogen and Antigravity Orchestrator (12 November 2026); Economic Times — Google Rolls Out Gemini 4 Hydrogen (13 November 2026).

Method: every factual claim in this article was verified on 13 November 2026 against Google’s own announcement, read in full, plus independent coverage of the launch. Benchmarks, speed figures and safety numbers are Google’s own and are labelled as such throughout; the Artificial Analysis Intelligence Index is an independent leaderboard. The worked cost example is our own arithmetic from Google’s published prices, with its token assumptions stated in the article. No hands-on testing was possible: the model launched on 12 November 2026 and the article was desk-researched.

Leave a Reply

Your email address will not be published. Required fields are marked *