Gemini 4 Hydrogen Is Out: In Antigravity on Google AI Pro, With a New Extra Thinking Level

By Lesia Kapliuk · Published November 12, 2026 · Updated November 13, 2026

TL;DR: On November 12, 2026, Google released Gemini 4 Hydrogen, the fast Gemini 4 model that Google recommends as the upgrade from Gemini 3.8 Flash. In Google Antigravity it reached our Google AI Pro account the same day, in the app, the IDE and the CLI, with no update needed. It has four thinking levels: Low, Medium, High and a new Extra (xhigh in the CLI and the API). Google also added Orchestrator, a model option in which a bigger model plans and Hydrogen does the work: Orchestrator (Argon) on Google AI Pro and Ultra, Orchestrator (Carbon) on Ultra only. In the Gemini API, gemini-4-hydrogen is generally available on the free tier and at $1.50 per million input tokens and $7.50 per million output tokens on the paid tier.

What Google announced

The Antigravity blog post from November 12 introduces two things: Gemini 4 Hydrogen as a model of its own, and Orchestrator, which pairs Hydrogen with Gemini 4 Carbon in one conversation. Google describes Hydrogen as built for "fast, dependable execution" in everyday coding and agent work, and Google AI Studio now greets visitors with a "Build with Gemini 4 Hydrogen" banner that calls it "our daily driver for rapid coding and dependable agentic execution".

Google's own numbers, which we have not checked:

Artificial Analysis Intelligence Index: 51 for Hydrogen, above Argon's 50, and 54 at its highest setting.
Terminal-Bench 4.1: 57%, against 18.2% for Gemini 3.8 Flash.
Orchestrator: near-Carbon task success on Terminal-Bench 4.1 while using on average 24% less quota and finishing 19% faster than Carbon alone.

The What's new in Gemini 4 Hydrogen guide adds that Hydrogen "cuts fabricated claims by more than 5x" compared with 3.8 Flash by flagging missing context instead of guessing, and that it batches read and search tool calls in parallel.

The name was not new. "Hydrogen" had been circulating for about a week as an unreleased model people met on LM Arena (example on Reddit); November 12 is the official release.

Who gets Gemini 4 Hydrogen in Antigravity

Google's two official pages don't say quite the same thing:

The blog post says Hydrogen "is rolling out across paid Google AI Pro and Ultra subscribers". Enterprises get it through Gemini for business.
The plan table on the models page marks Hydrogen as available on every plan, including Free and Google AI Plus:

Model	Free & Google AI Plus	Google AI Pro	Google AI Ultra	Enterprise
Gemini 4 Hydrogen	yes	yes	yes	yes
Orchestrator (Carbon)	no	no	yes	no
Orchestrator (Argon)	no	yes (paid plans)	yes	no
Gemini 4 Carbon	no	no	yes	yes
Gemini 4 Argon	no	yes (paid plans)	yes	yes
Gemini 3.8 Flash	yes	yes	yes	yes

As shown on the models page on November 13, 2026. On November 12 the page and the blog post still listed a single Orchestrator for Ultra only.

We can confirm only the Pro part: our paid Google AI Pro account has Hydrogen. If you are on a free plan and don't see it yet, the blog post's wording suggests free plans may come later; the changelog also says new versions "roll out gradually and may take a few days to reach all users".

What a Google AI Pro account sees

Hydrogen appeared in our model list on the evening of November 12 (Kyiv time), in the app, the IDE and the CLI. No update was involved: the model list comes from Google's servers, so the Antigravity app 2.31.2 and CLI 1.5.7 we already had simply showed the new entry.

In the app, Hydrogen sits between Argon and 3.8 Flash with a "New" tag. Its submenu has four levels, Low, Medium, High and Extra:

In the CLI, /model shows the same levels on an Effort slider, where the fourth one is called xhigh:

agy models now lists four entries for it:

gemini-4-hydrogen-high     Gemini 4 Hydrogen (High)
gemini-4-hydrogen-xhigh    Gemini 4 Hydrogen (XHigh)
gemini-4-hydrogen-medium   Gemini 4 Hydrogen (Medium)
gemini-4-hydrogen-low      Gemini 4 Hydrogen (Low)

A few details we checked ourselves:

Extra and xhigh are the same level. The models page says so directly: Extra in the selector sets the "Extra High" reasoning budget, xhigh in the CLI.
Two ways to pick it in the CLI. --model gemini-4-hydrogen-xhigh and --model gemini-4-hydrogen --effort xhigh both work. Mixing a level in the name with a different --effort (for example gemini-4-hydrogen-xhigh --effort max) is rejected as a conflict.
No max for Hydrogen. The CLI's --effort accepts max, but for Hydrogen it answers that the model has no "max" effort and lists low, medium, high and xhigh.
Hydrogen shares a usage pool with Gemini 3.8 Flash. In the usage data our Pro account receives, all four Hydrogen levels show the same remaining share and the same reset time as 3.8 Flash; Argon and the Claude models have pools of their own. Our guess is that Antigravity counts Hydrogen against the Flash allowance. Google doesn't say how much of it a Hydrogen request uses compared with 3.8 Flash.

Orchestrator: Argon on Pro, Carbon on Ultra

Orchestrator is a separate entry in the model selector, not a thinking level. According to the models page, a bigger "planner" model explores, plans and reviews; Hydrogen applies the edits, runs builds and tests and checks the result against the plan; when Hydrogen runs into trouble (Google's example is a verification step that fails twice) it hands control back to the planner, which repairs the plan. Google stresses that this is not a router that classifies your first prompt: the hand-offs happen during the task.

There are two versions. Orchestrator (Carbon) plans with Gemini 4 Carbon and is for Google AI Ultra. Orchestrator (Argon) plans with Gemini 4 Argon and is available on paid Google AI Pro as well as Ultra. Google launched Orchestrator on November 12 as Ultra-only; the Argon version and the Pro access were added to the models page overnight.

On our Pro account, Orchestrator (Argon) appeared in the model list between 20:36 and 23:50 UTC on November 12, with the tooltip "Argon that may offload routine steps to Hydrogen. Same Argon-level quality, fewer tokens used." In our usage data it shares the remaining share and reset time of Argon, not of 3.8 Flash and Hydrogen. In the CLI it shows up in agy models as gemini-pro-orchestrator-g1-pro, and the models page says to select it by name: --model "Orchestrator (Argon)".

Orchestrator runs with a fixed configuration, so you can't choose its thinking level, and --effort doesn't apply to it. Google calls it the recommended default for most day-to-day development on the plans that have it, and suggests picking Hydrogen directly for routine work such as package upgrades, boilerplate, unit tests and quick scripts.

Gemini 4 Hydrogen in the Gemini API

Outside Antigravity, Hydrogen is generally available in Google AI Studio and the Gemini API, on both the free and the paid tier, according to the AI Studio changelog. The model ID is gemini-4-hydrogen, and it appeared in the API's model list on our check that evening.

	Gemini 4 Hydrogen
Context window	1,048,576 tokens, shared by input, thinking and output
Output	65,536 tokens per request by default; up to 1M with long decode
Input	text, image, video, audio, PDF
Thinking levels	low, medium, high (default), xhigh
Paid tier	$1.50 / 1M input tokens, $7.50 / 1M output tokens (thinking included)
Free tier	yes
Context caching	implicit only (no explicit caching)

Source: AI Studio pricing, model page and What's new in Gemini 4 Hydrogen, read November 12, 2026.

Things to know if you are moving from gemini-3.8-flash:

The price is double, for now. 3.8 Flash costs $0.75 / $3.75 per million tokens through December 31, 2026 and $1.50 / $7.50 from January 1, 2027 (pricing). So Hydrogen costs what 3.8 Flash will cost next year, and twice what it costs today.
The default thinking level is higher. Hydrogen defaults to high (3.8 Flash defaults to medium), and minimal is rejected with an error. Google warns that xhigh can overthink simple prompts and use up the output limit, and advises testing all four levels with a high max_output_tokens.
No explicit caching. Requests with cached_content have to change. Implicit caching works on prompts of at least 4,096 tokens and, according to Google, discounts cached input tokens by 95%.
The free tier is small. Our free-tier key was cut off after 20 generateContent requests to Hydrogen in a day; the error named a per-model daily limit of 20. That is enough to try it, not to run an evaluation.
Managed Agents: the Antigravity agent in the Gemini API now runs on Hydrogen by default.

What this means for you

On Google AI Pro: Hydrogen should be in your model selector now. It looks like the new everyday model in place of 3.8 Flash, and it draws from the same allowance (about 1.6 times as fast as Flash in our test), so try Medium or High first and keep Extra for hard problems. Argon is still there for long, difficult tasks, and the new Orchestrator (Argon) lets Argon plan while Hydrogen does the routine steps, on the Argon allowance.
On Google AI Ultra: you get Hydrogen and both Orchestrators. Google suggests Orchestrator as the default for everyday development and Carbon alone for the hardest work.
On a free plan: the models page lists Hydrogen for you, the blog post mentions only paid Pro and Ultra. If it isn't in your selector yet, it isn't a problem on your side.
If you use the Gemini API: Hydrogen is the recommended upgrade from 3.8 Flash, but until the end of the year it costs twice as much per token. Remove explicit caching and low max_output_tokens caps before switching.
If you use the Antigravity CLI with an API key: CLI 1.5.8, released overnight on November 13, adds Hydrogen (and Carbon) to /model and --model when you connect with a GEMINI_API_KEY, and Hydrogen when you sign in with Application Default Credentials.

How to check it yourself

In the Antigravity CLI, run agy models, or start agy, type /model and move the Effort slider with the arrow keys.
In the Antigravity app, open the model selector under the prompt box and hover over Gemini 4 Hydrogen to see its levels.
Compare with the plan table on Google's models page.

What we don't know

When, or whether, free and Google AI Plus accounts get Hydrogen in Antigravity, given the difference between the blog post and the models page.
How much more of the shared allowance Extra uses than High. (On Medium, Hydrogen used about 1.6 times the allowance of 3.8 Flash for the same work in our test.)
How much Argon allowance Orchestrator (Argon) saves compared with running Argon alone. Google says "fewer tokens used" but gives no figure for the Argon version.

We have since tested all four levels against Gemini 3.8 Flash: Gemini 4 Hydrogen vs 3.8 Flash: thinking levels, speed and quota.

Update history

Nov 13, 2026: Orchestrator (Argon) reached Google AI Pro, and Orchestrator split into Argon and Carbon versions; Antigravity CLI 1.5.8 added Hydrogen for API-key and Application Default Credentials sign-in. Text updated.
Nov 13, 2026: linked our thinking-level and quota test.