Gemini 4 Hydrogen: What Google Launched, What It Costs and Who Should Use It

Until this week, “Gemini 4 Hydrogen” was a codename passed around on X. On November 12, 2026 it became a shipping product, and most of what was written about it in the previous two weeks is now out of date. This article reflects what is known as of November 13, 2026: what Google released, the numbers it published, what independent testers measured, and where the model makes sense.

The short version for developers: Hydrogen takes over from Gemini 3.8 Flash as Google’s everyday model, thinks harder by default, finishes agentic tasks much faster, and costs twice what Flash’s introductory price does. Whether that trade pays off depends on your workload, and the sections below are built to help you decide.

Contents

• What Is Gemini 4 Hydrogen?
• From Leak to Launch
• Specs, Model ID and Thinking Levels
• Pricing
• Benchmarks: Vendor Claims vs Independent Tests
• Gemini 4 Hydrogen vs 3.8 Flash, Argon and Carbon
• Where to Use It
• Migrating From Gemini 3.8 Flash
• Where It Falls Short
• Is It Worth Switching?
• FAQs
• Sources

What Is Gemini 4 Hydrogen?

Gemini 4 Hydrogen is the third model in Google’s Gemini 4 family and the first one built for speed and cost rather than top-end reasoning. Gemini 4 Argon arrived on September 30 and Gemini 4 Carbon followed on October 28. Both aim at the hardest long-horizon work. Hydrogen handles the volume: code edits, document processing, tool calls and everyday search answers.

Google dropped the Flash name for it. Hydrogen sits in the slot Gemini 3.8 Flash held, and Google’s developer guidance, as quoted by TeamoRouter, calls it the recommended upgrade from 3.8 Flash and earlier Gemini 3 and 2.5 models. XDA points out that Google now appears to be naming models after chemical elements. The name has nothing to do with hydrogen fuel or energy.

How Gemini 4 Hydrogen Went From Leak to Launch

The Hydrogen story ran fast. Google shipped three Gemini 4 models in six weeks, after a long stretch in which it released only smaller Flash models while rivals moved ahead, a gap Axios described at the Argon announcement.

The leakers got the name right. The speed figure is a different matter. Nothing Google published matches “300 to 500 tokens per second”. Its own claim is about finishing tasks: 2.5 times faster than Gemini 3.8 Flash at Hydrogen’s default reasoning setting. Those are different measurements, and a model can generate tokens slowly yet complete a job quickly if it needs fewer retries.

Gemini 4 Hydrogen Specs, Model ID and Thinking Levels

Hydrogen shares its architecture family, its one-million-token window and its built-in tools with Carbon and Argon. What separates it is how long it deliberates. It exposes four thinking levels, and xhigh is new with this model. Google’s guidance, relayed by TeamoRouter, is to start at high and reserve xhigh for hard multi-step problems where high falls short, because xhigh tends to overthink simple prompts.

A few capabilities are missing. According to the official model page as summarized by TeamoRouter, Hydrogen has no image generation, no Live API and no audio generation. Explicit context caching is not supported, only implicit caching. VentureBeat adds that Google’s migration guide rejects the old thinking_budget parameter and the MINIMAL setting, and that automatic caching on Google Cloud starts at 4,096 tokens, down from 6,144 on Flash. Output defaults to 65,536 tokens per foreground segment and can run longer through long decode.

Gemini 4 Hydrogen Pricing: What You Actually Pay

Standard rates are $1.50 per million input tokens and $7.50 per million output tokens, with reasoning tokens billed as output. That is exactly twice Gemini 3.8 Flash’s introductory rate, which runs through December 31, 2026, and it matches the rate Flash moves to on January 1, 2027, according to VentureBeat. Google lists no introductory discount for Hydrogen.

Standard, Batch and Flex rates are confirmed by VentureBeat and TeamoRouter; the Priority row comes from TeamoRouter’s summary of Google’s pricing page. TeamoRouter also reports a free tier in the Gemini API.

Per-token price is only half of the cost question, because Hydrogen and Flash spend different numbers of tokens per task. Artificial Analysis estimates what a full benchmark task costs, and the results cut against intuition: Flash at high reasoning, despite its lower token price, comes out at $7.77 per task, far above Hydrogen at the same level.

Source: Artificial Analysis figures as reported by VentureBeat. They are estimates for one benchmark suite, not a quote for your workload.

Two readings follow. Hydrogen at medium scores above Haiku 5.5 at max in this index, at roughly double Haiku’s estimated cost per task. And the cheap end of the market is still far cheaper: VentureBeat calculates that Hydrogen costs 15 times the lowest standard per-token rates of Haiku 5.5, GPT-6 Luna and Meta’s Muse Ember.

Gemini 4 Hydrogen Benchmarks: Vendor Claims vs Independent Tests

Keep the sources apart when reading Hydrogen’s numbers. Google’s launch figures are self-reported, and benchmark versions differ: Terminal-Bench 4.1 in Google’s table is not the Terminal-Bench 4.0 that Artificial Analysis ran.

Terminal-Bench comparisons have wobbled before. Coverage of Argon’s launch noted that Google used a rival’s self-reported 66.4% where Artificial Analysis measured 59.6% under a different setup, which is why the independent column matters more than the launch chart. Hydrogen’s 54% matched at two settings is also worth noticing: moving from high to xhigh raised the overall index by three points and the estimated task cost from $1.87 to $3.16, with no gain on this coding test.

Gemini 4 Hydrogen vs Gemini 3.8 Flash, Argon and Carbon

Google’s own suggestion for combining them is practical: let Carbon plan system interfaces and clear hard roadblocks, then hand code edits, parallel subagent tasks and test runs to Hydrogen. Antigravity’s new Orchestrator can switch between the two in one session, and the Gemini Managed Agents’ Antigravity agent defaults to Hydrogen, per TeamoRouter.

Where to Use Gemini 4 Hydrogen Today

Plan details come from Notebookcheck, which also notes that Google’s plans page and its help page currently disagree about which models Plus subscribers keep. If you rely on a specific tier, check Google’s pages before paying. In the Gemini app, you can see which model wrote a reply by opening the menu under a response, choosing Show thinking steps and scrolling to the bottom.

Moving From Gemini 3.8 Flash to Gemini 4 Hydrogen

1. Change the model string. Use gemini-4-hydrogen. Older leak coverage used gemini-4-flash, which is not the shipping ID.
2. Remove deprecated reasoning controls. The thinking_budget parameter and the MINIMAL setting are rejected, so switch to the four named levels.
3. Expect a bigger bill at the default. Hydrogen defaults to high, where Flash defaulted to medium. Test medium before accepting the default; in Artificial Analysis’s suite it costs less than half as much per task as high.
4. Rework caching. Explicit caching is not available. You get implicit caching only, so repeated prompts save money only when the cache hits.
5. Run your own evaluation at all four levels. Give outputs enough room that the token cap does not truncate answers, then pick the lowest level that meets your quality bar.
6. Check the calendar before rushing. Flash’s introductory price lasts through December 31, 2026, and Google’s deprecation schedule lists no shutdown date for it.

Where Gemini 4 Hydrogen Falls Short

• It is not the cheap option. Several rival small models cost a fraction per task, and Xiaomi’s hosted MiMo-V2.6-Flash is estimated at $0.13 per benchmark task against $1.87 for Hydrogen at high, with a 24-point gap in the index.
• Coding is competitive, not dominant. Google’s own table puts Claude Sonnet 5.5 ahead on Terminal-Bench 4.1.
• Feature gaps. No image generation, no Live API and no explicit caching.
• Thin independent evidence so far. The model is one day old. Most numbers come from Google or from a single benchmark provider, and customer results such as Box’s are early-access reports.
• Unpublished details. A knowledge cutoff for Hydrogen was not in the sources reviewed here. Argon’s model card lists March 2026, but that does not confirm Hydrogen’s.

Is Gemini 4 Hydrogen Worth Switching To?

It is worth a test if you run Flash for agentic coding, tool-heavy workflows or document and table work and you are paying for retries, slow runs or wrong answers. The reported gains on those jobs are large enough that a per-task comparison could favour Hydrogen despite the higher token price. If your workload is high-volume classification, extraction or short chat, a cheaper small model, or Flash through the end of its introductory pricing, will likely cost less for the same output. Measure on your own prompts before changing anything in production.

Gemini 4 Hydrogen FAQs

Is Gemini 4 Hydrogen the same as Gemini 4 Flash?

Yes, in practice. The Flash-tier model leakers called “Hydrogen” shipped under that name, and Google dropped the Flash label. The API model ID is gemini-4-hydrogen.

Is Gemini 4 Hydrogen free?

There is a free tier in the Gemini API, according to TeamoRouter’s reading of Google’s pricing page, and free Gemini app users reach the model only through auto model selection. Paid API use starts at $1.50 per million input tokens and $7.50 per million output tokens.

Can Gemini 4 Hydrogen generate images?

No. The official model page lists image generation, audio generation and the Live API as unsupported. It accepts images, video, audio and PDFs as input and returns text.

Is Gemini 4 Hydrogen better than Gemini 4 Argon?

On Artificial Analysis’s Intelligence Index at high reasoning, Hydrogen scores 51 to Argon’s 50, so the gap is one point. That does not make it better at everything. Argon is built for long-horizon professional and security work and carries a 1-million-token output limit.

Which thinking level should I use?

Start with high, the default, and compare against medium. Use xhigh only if high fails on hard multi-step tasks. Independent testing showed xhigh adding index points at higher cost but no extra Terminal-Bench 4.0 score.

Why is it called Hydrogen?

Google has not explained the choice in the sources reviewed here. Argon, Carbon and Hydrogen are all chemical elements, and the family appears to follow that pattern. It is unrelated to hydrogen fuel or fuel cells.

Is Gemini 3.8 Flash being shut down?

Not yet. Google’s deprecation schedule lists no shutdown date for Flash, and its introductory pricing runs through December 31, 2026.

Sources and References

• VentureBeat: Google launches Gemini 4 Hydrogen, a faster, smarter, pricier Flash successor (Nov 12, 2026)
• Notebookcheck: Gemini 4 Hydrogen, what free users and Google AI Plus get (Nov 13, 2026)
• TeamoRouter: Gemini 4 Hydrogen API, price, thinking levels (Nov 13, 2026; summarizes Google’s model and pricing pages)
• OfficeChai: Google preparing to release a Flash model named Gemini 4 Hydrogen (Nov 5, 2026)
• XDA Developers: Gemini 4 Hydrogen replaces Flash in Search
• Google: Introducing Gemini 4 Hydrogen (cited by VentureBeat and Notebookcheck)
• Axios: Google unveils long-awaited Gemini 4 (Sep 30, 2026)
• ComputingForGeeks: Gemini 4 Argon pricing, benchmarks and API
• Khaleej Times (Reuters): Google launches Gemini 4 Carbon
• Trending Topics: Gemini 4 Argon, Google, Anthropic and OpenAI
• AGTP on X: summary of Business Insider report (Nov 11, 2026) (second-hand; original not reviewed)