Gemini 4 Hydrogen Is Live: $1.50 API, Free Tier, Beats Argon

Google has a new default model for everyday AI work, and it arrived with a price cut attached. Gemini 4 Hydrogen launched on November 12, 2026, and it does two things the rest of the Gemini 4 family does not: it costs $1.50 per million input tokens, and it has a free API tier. The model is built for coding and multi-step agent work rather than the deep, long-horizon reasoning that Gemini 4 Carbon and Gemini 4 Argon handle, and Google says it still scores higher than Argon on an independent intelligence ranking.

That combination is the story. Hydrogen is cheaper than both of its bigger siblings, faster than the model it replaces, and available in the free Gemini app and Google Search from day one. The trade-offs sit in the fine print: thinking tokens are billed, the free tier comes with a data-use condition, and the headline benchmark numbers are Google's own. Here is what Hydrogen is, what it costs in the US, UK and Canada, and where it fits.

What Gemini 4 Hydrogen is

Hydrogen is the fast, lower-cost member of the Gemini 4 family. Google describes Carbon and Argon as its frontier models for deep reasoning, and Hydrogen as the model for rapid execution across everyday coding and agentic workflows. It is adapted from Gemini 4 Carbon, according to Unite.AI, and tuned for the jobs that fill a working day: writing code, running tests, fixing bugs, reading documents, and chaining tools together in multi-step tasks.

The announcement, published November 12 by Koray Kavukcuoglu, Google DeepMind's senior vice president and chief AI architect, puts Hydrogen on every surface at once. It takes in text, images, audio and video, with a knowledge cutoff of March 2026, according to Unite.AI. It has a 1 million-token input window and can produce up to 1 million output tokens. It takes over from Gemini 3.8 Flash as the default workhorse model in Google's lineup, and it is the third Gemini 4 model to ship in about six weeks, after Argon and Carbon.

Google's pitch is speed with discipline. The company says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting, scores 57% on Terminal-Bench 4.1 against 18.2% for Flash, and produces fabricated claims more than five times less often. Those are Google's own test results, not independent measurements, and they should be read that way.

What Gemini 4 Hydrogen costs

The API price is flat across the whole 1 million-token context window: $1.50 per million input tokens and $7.50 per million output tokens. Cached input costs $0.075 per million tokens through implicit caching. Every token the model processes is counted and billed, including thinking tokens and internal tool instructions, according to Google's pricing page.

That flat rate is the point. Carbon and Argon charge $2 per million input tokens and $10 per million output tokens through November 30, 2026, then double to $4 and $20 from December 1. Hydrogen is 25% cheaper than their current promotional rate and 62.5% cheaper than the rate that starts in December, as Neowin calculated. Against Gemini 3.8 Flash, which costs $0.75 and $3.75 through December 31, Hydrogen is twice the price, but Flash rises to $1.50 and $7.50 on January 1, 2027, exactly what Hydrogen costs today.

At November 13, 2026 mid-market rates (1 USD = 0.7557 GBP, 1.4256 CAD), the standard rate works out to about £1.13 and C$2.14 per million input tokens, and about £5.67 and C$10.69 per million output tokens. Those are approximations; Google bills in US dollars.

The free tier is the detail that separates Hydrogen from its siblings. Carbon and Argon have no free tier at all. Hydrogen's input, output and cached tokens are free of charge on the free tier, with rate limits, according to Google's pricing page and Unite.AI. The condition is in the terms: on the free tier, content submitted to the API may be used to improve Google's products. Paid-tier content is not used that way. Grounding with Google Search is also not available on the free tier. On paid tiers, the first 5,000 search requests a month are free, then $14 per 1,000 requests, with a daily cap of 1,500 requests.

Batch and Flex modes cut the standard rate in half, to $0.75 and $3.75, for work that can wait or run at lower priority. Priority mode costs more for faster handling: $2.70 and $13.50. The price is the same no matter how much of the context window a request uses, which is the part that matters for agent loops that fire hundreds of small requests per task. Unlike Carbon and Argon, Hydrogen's standard rate carries no introductory promotion and no published expiry date, according to VentureBeat. The $1.50 and $7.50 rates are the standing price, not a discount that ends on a set date.

How Hydrogen compares

The headline comparison is against Argon. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high thinking level against 50 for Argon, and reaches 54 at its highest xhigh setting, according to Google, citing Artificial Analysis. The Decoder, reporting on the same Artificial Analysis data, puts Carbon at 59, and Claude Sonnet 5.5 and OpenAI's GPT-6.1 Sol at 55 each at their maximum settings. At its default level, Hydrogen sits slightly behind GPT-6.1 Sol at 52 and Claude Opus 5.5 at 52.

* Carbon and Argon prices are the promotional rates through November 30, 2026; both rise to $4 and $20 from December 1. ** Gemini 3.8 Flash prices run through December 31, 2026, then rise to $1.50 and $7.50. Index scores are from Artificial Analysis, as reported by The Decoder and VentureBeat; GPT-6.1 Terra pricing is per Neowin.

The one-point gap over Argon is small, and the price gap is the bigger story. Hydrogen costs 25% less than Argon's current rate while scoring a point higher, and it is 50% more expensive than OpenAI's GPT-6.1 Terra at $1 and $5, according to Neowin. On cost per task, Artificial Analysis puts Hydrogen at about $1.87 per benchmark task at high and $3.16 at xhigh, per VentureBeat, which is the cost of the extra reasoning the xhigh setting buys.

Hydrogen is not the cheapest model in its class, and the gap is wide. Anthropic's Claude Haiku 5.5, OpenAI's GPT-6 Luna and Meta's Muse Ember all charge $0.10 per million input tokens and $0.50 per million output tokens at their lower tiers, according to VentureBeat, roughly 15 times less than Hydrogen per token. The trade is capability: those three score 36, 28 and 36 on the Artificial Analysis Index at their maximum settings, against 51 for Hydrogen at high. An early-access test at Box, cited by VentureBeat, moved accuracy from 74% with Flash to 79% with Hydrogen on the company's Complex Work Eval, in less than half the time and with about 12% fewer tokens.

Terminal work is the weak spot. In Artificial Analysis's Terminal-Bench 4.0 runs, Hydrogen scored 54%, ahead of Meta's Muse Ember at 40.9%, OpenAI's Luna at 12.6% and Anthropic's Haiku at 32.8%, according to VentureBeat, but The Decoder notes Hydrogen scored only 50% on Terminal-Bench-Science 0.1. Google's own Terminal-Bench 4.1 figure of 57% is a different, newer test, so the two numbers should not be read side by side.

On other Google-reported benchmarks, 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%, ranks fifth on Agent Arena, and scores 53.4% on the Multimodal Knowledge Agent benchmark and 71.6% on Agents' Last Exam v1.1. All of those are Google's own figures, and none has an independent check attached yet.

Four thinking levels

Hydrogen offers four thinking levels: low, medium, high, which is the default, and a new xhigh tier. Google's framing is that the levels scale in a smooth progression, so you only wait and pay for the reasoning depth a task needs. Argon, by comparison, offers only medium and high, according to The Decoder.

The xhigh setting is where the trade-off shows. It lifts the Artificial Analysis score from 51 to 54, but it uses nearly twice as many output tokens as high, according to The Decoder, and output tokens are the expensive half of the bill at $7.50 per million. That makes xhigh a setting for the rare request that really needs it, not a default. The lower levels still hold up: medium scores 41 and low scores 27 on the same index, per VentureBeat, and even at medium Hydrogen outscores Haiku 5.5, GPT-6 Luna and Muse Ember at their maximum settings. One more detail from Google's documentation, noted by CellCog: Hydrogen does not accept the minimal thinking level that some other Gemini models offer, and the API returns an error if you ask for it.

Where you can use it

Hydrogen is rolling out globally across Google's developer, enterprise and consumer surfaces. Developers get it through the Gemini API on all tiers, including the free tier, and in AI Studio and Android Studio. Enterprises get it in Gemini for business and in Gemini in Google Sheets for eligible subscriptions. For everyone else, it appears in the Gemini app model dropdown and in Google Search for all users globally, while free Gemini app users receive it through automatic model selection, according to Unite.AI.

Paid Google AI Pro and Ultra subscribers get Hydrogen in Google Antigravity. Google AI Pro costs $19.99 a month, and Ultra starts at $99.99 a month, according to Google's subscriptions page. Antigravity also gains Orchestrator, a new model dropdown option that switches between Carbon and Hydrogen inside the same session. Google says Orchestrator reached near-Carbon task success rates on Terminal-Bench 4.0 while using 24% less quota and finishing 19% faster. The idea is that Carbon acts as the coordinator while Hydrogen powers the worker and verifier agents, which is a cheaper way to run parallel agents than putting Carbon on every seat.

Google also launched a different kind of model this month: the Gemini Decisions API, a classification model that charges $0.10 per million input tokens with no output charge at all. It is a separate product for a separate job, and it shows how fast Google is splitting its model lineup by workload rather than by size alone.

The catches

Hydrogen is a strong launch, but five details deserve a clear read before anyone switches a workload to it.

First, the free tier is not free of conditions. Content sent on the free tier may be used to improve Google's products, according to the pricing page. Anyone handling client data, unpublished work or anything sensitive should use the paid tier, where that does not happen.

Second, thinking tokens are billed. The pricing page states that every token the model processes is counted, including thinking tokens and internal tool instructions. The xhigh setting's extra reasoning is real money: Artificial Analysis puts the cost per benchmark task at $3.16 at xhigh against $1.87 at high, per VentureBeat.

Third, the benchmark story is mostly Google telling you about Google. The Artificial Analysis Intelligence Index is independent, but the one-point lead over Argon is thin, and the Terminal-Bench, HLE-Diamond, Vals and Agent Arena figures all come from Google's own announcement. Independent testing had only just begun at publication time.

Fourth, the knowledge cutoff is March 2026, according to Unite.AI. For a model shipping in November, that is eight months of news, releases and price changes it has not seen, which matters for research tasks and for any question about a product launched this year.

Fifth, Hydrogen is closed. Google has not released the model weights and has not disclosed the model size or parameter counts, according to SiliconReport. That rules out self-hosting and local deployment, and it means the free tier is the only way to run it without paying Google.

Who should use it

Hydrogen fits developers running agent loops and coding assistants, teams migrating from Gemini 3.8 Flash, and anyone who wants Gemini 4 quality at a lower per-token rate with a free tier to test against. The 1 million-token window and the batch discount make it a practical choice for large-repo work and document-heavy agent tasks, where the input side of the bill usually dominates.

It is not the pick for the hardest reasoning jobs. Google's own framing leaves those to Carbon and Argon, and the Artificial Analysis table agrees: Carbon at 59 sits well above Hydrogen's 54 at xhigh. If a task needs the deepest available reasoning, or if a workflow cannot tolerate the free tier's data-use terms, the paid Carbon or Argon tiers are the honest choice, at their higher price.

How to try it

1. Open Google AI Studio, start a new chat, and select gemini-4-hydrogen from the model list. No payment method is needed for the free tier.
2. For API use, create a key in AI Studio and call the model as gemini-4-hydrogen. The free tier covers input, output and cached tokens, with rate limits.
3. In the Gemini app, pick Hydrogen from the model dropdown. Free users receive it through automatic model selection.
4. Google AI Pro and Ultra subscribers can also try it in Antigravity, including the Orchestrator option that pairs it with Carbon.

Frequently asked questions

How much does Gemini 4 Hydrogen cost?

The standard API rate is $1.50 per million input tokens and $7.50 per million output tokens, flat across the 1 million-token context window. Cached input costs $0.075 per million tokens. Batch and Flex modes halve the rate to $0.75 and $3.75, and Priority mode costs $2.70 and $13.50.

Is Gemini 4 Hydrogen free?

Yes, in two ways. The Gemini API has a free tier where Hydrogen input, output and cached tokens are free of charge, with rate limits, and the model is also rolling out in the free Gemini app and in Google Search. The free API tier has one condition: content sent on it may be used to improve Google's products.

Is Hydrogen better than Gemini 4 Argon?

On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high setting against 50 for Argon, and 54 at its highest xhigh setting. It is also 25% cheaper than Argon's current $2 and $10 rate. The gap is small, and Argon and Carbon remain Google's frontier picks for the deepest reasoning work.

What replaced Gemini 3.8 Flash?

Hydrogen takes over as the default workhorse model in Google's lineup. Gemini 3.8 Flash is still listed on the pricing page at $0.75 and $3.75 through December 31, 2026, then rises to $1.50 and $7.50 on January 1, 2027, the same as Hydrogen costs today.

What are the four thinking levels?

Low, medium, high and a new xhigh tier. High is the default. The levels trade reasoning depth against latency and cost, so routine work runs on low or medium and only the hardest tasks use xhigh, which nearly doubles output-token use.

Where is Gemini 4 Hydrogen available?

Through the Gemini API on all tiers, in AI Studio and Android Studio, in Gemini for business, in the Gemini app and Google Search, and in Antigravity for paid Google AI Pro and Ultra subscribers. It is rolling out globally from November 12, 2026.

Sources

• Google, "Introducing Gemini 4 Hydrogen for Agentic Workflows," November 12, 2026, blog.google
• Google, "Gemini Developer API pricing", ai.google.dev
• Google Antigravity, "Gemini 4 Hydrogen and Orchestrator," November 12, 2026, antigravity.google
• The Decoder, "Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch," November 12, 2026, the-decoder.com
• VentureBeat, "Google launches Gemini 4 Hydrogen — a faster, smarter, pricier Flash successor," November 12, 2026, venturebeat.com
• Neowin, "Google's Gemini 4 Hydrogen beats GPT-6.1 Terra on most key AI benchmarks," November 12, 2026, neowin.net
• Unite.AI, "Google Launches Gemini 4 Hydrogen for Everyday Agentic Workflows," November 12, 2026, unite.ai
• MadRobot, "Gemini 4 Hydrogen: price, free access and the Search upgrade," November 12, 2026, madrobot.blog
• CellCog, "Gemini 4 Hydrogen Released: Price, Benchmarks and Access," November 12, 2026, cellcog.ai
• SiliconReport, "Google launches Gemini 4 Hydrogen, replacing Gemini 3.8 Flash," November 12, 2026, siliconreport.com
• Google, "Google AI Pro and Ultra" subscriptions page, gemini.google
• AI Core Updates, "Gemini 4 Argon: $2/$10 Through Nov 30, Then It Doubles", aicoreupdates.com (internal)
• AI Core Updates, "Gemini Decisions API: 1,000 Questions for $0.10, No Output Charge", aicoreupdates.com (internal)

Exchange rates: November 13, 2026 mid-market rates, verified at publication (1 USD = 0.7557 GBP, 1.4256 CAD). Prices are in US dollars unless stated otherwise.