Gemini 4 Hydrogen Is Out! Can AI Pro Plan Users Finally Find Peace?

3-Line TL;DR

Google launched Gemini 4 Hydrogen on November 12, and it scores 54 at xhigh (51 at default high) on the Artificial Analysis index at 188 tok/s, beating Argon's 50 at 65 tok/s
Carbon is still unavailable on AI Pro, while Hydrogen opens to Free and AI Pro users too
Pro's five-hour and weekly caps stay, but cheap, fast Hydrogen could be the relief pitcher (author's view)

Google released Gemini 4 Hydrogen (gemini-4-hydrogen) on November 12 with four thinking levels: low, medium, high and xhigh (Google, AI Studio models)
On Artificial Analysis (AA) Intelligence Index v5.0.1, default high scores 51 and xhigh reaches 54, ahead of Gemini 4 Argon's 50 (AA Hydrogen, AA Argon)
Lower levels fall off fast: medium scores 41 and low 27, both below Argon
Output speed jumps to 179 tok/s at high and 188 tok/s at xhigh, roughly 2.7 times Argon's 65 tok/s
Time per AA index task drops from 30.0 minutes on Argon to 7.9 minutes on Hydrogen high

Bars: AA index Hydrogen high 51, xhigh 54, Argon 50, Carbon 60; speed Hydrogen 179–188 tok/s vs Argon 65 and Carbon 72

Paid API rates sit at $1.50 input and $7.50 output per 1M tokens across a 1M context window (AI Studio pricing)
Argon and Carbon cost $2 and $10 through November 30, then double to $4 and $20 on December 1
On AA's full index run, Hydrogen xhigh costs $3.16 per task versus $4.97 for Argon, though GPT-6.1 Sol max is still cheaper at $1.27
Google claims Hydrogen cuts fabricated claims by more than 5x versus Gemini 3.8 Flash, while independent AA-Omniscience hallucination percentages could not be read from AA's client-rendered charts
In my own use, hallucinations are clearly down and, above all, it is the fastest, which I like (author's impression, not a measurement)

In our October Carbon post, Carbon was limited to cyber defenders and AI Pro users already felt cramped on Argon
Carbon opened broadly on November 10, but in the app and Antigravity only Google AI Ultra subscribers got it (Google Carbon)
The rest went to paying API developers and enterprises
Antigravity's model table still marks Gemini 4 Carbon and Orchestrator (Carbon) with red crosses on Google AI Pro (Antigravity docs)
Hydrogen and Orchestrator (Argon) both get green checks on paid AI Pro

Capture of Google Antigravity Reasoning model table with a red outline around Gemini 4 Carbon, Argon, Hydrogen and Orchestrator rows

Source: Google Antigravity Docs — Reasoning model plan table; red box added by HSL

Argon's allowance is tiny and Carbon is off-limits, so ordinary AI Pro ($19.99/month) users are loudly unhappy (author's view)
Honestly, it feels like hardly anyone pays for Gemini these days (author's view)
No data counts subscribers, and the launch HN thread does include people paying for Pro who hit the Argon limit (Hacker News)
Even so, an affordable, capable Hydrogen looks like the relief pitcher AI Pro needed (author's view)

Google still enforces compute-based five-hour and weekly limits on both the Gemini app and Antigravity, with no published prompt counts (Gemini Help, Antigravity plans)
Google has not said how much less of the limit Hydrogen uses than Argon
Since limits track compute, it should drain slower, but that is an estimate
Antigravity also added Orchestrator: a larger model plans and reviews, and Hydrogen does the edits and tests (Antigravity docs)

Orchestrator flow: Argon or Carbon plans, Hydrogen edits and tests, planner repairs the plan on friction

Made by HSL — Google Antigravity Docs, checked November 13, 2026

On Terminal-Bench 4.0, Google says Orchestrator came near Carbon's success rate while using 24% less quota and finishing 19% faster
That figure is for the Carbon plus Hydrogen pairing on Ultra
Pro only gets Orchestrator (Argon), and Google published no numbers for it

Official launch clip uploaded by Google Antigravity

Table: Hydrogen Free/Pro/Ultra at $1.50/$7.50, Argon Pro/Ultra and Carbon Ultra/Enterprise at $2/$10, $4/$20 from Dec 1

Made by HSL — Antigravity Docs and AI Studio Pricing, checked November 13, 2026

xhigh produced 304M output tokens across the AA run, so leaning on it will likely drain quotas faster (estimate)

Hacker News user threecheese asks whether Hydrogen in Antigravity works on a Pro plan or requires Ultra, after burning through Pro tokens on Argon in four minutes (no measurements) (Hacker News)
User data-ottawa reports on medium reasoning that Hydrogen cost 37% more per task than 3.8 Flash in their evals while finishing in 40–60% of the time (individual user test) (Hacker News)
User wds criticizes Google's element naming after seeing 3.8 Flash replaced by 4 Hydrogen, saying element names do not tell you which model is stronger (no measurements) (Hacker News)
User artdigital welcomes a fast model because Argon and Carbon feel very slow compared with Opus (no measurements) (Hacker News)

Q. Can AI Pro users finally use Gemini 4 Carbon?
- No. Carbon and Orchestrator (Carbon) work only on Ultra, Enterprise and the paid API
Q. Is Hydrogen actually better than Argon on Pro?
- At high or above it beats Argon on the AA index and runs about 2.7x faster
- At medium or below it scores lower than Argon
Q. Does Hydrogen fix the Pro usage limit problem completely?
- Not completely. The five-hour and weekly compute caps remain, and running Hydrogen at xhigh burns far more tokens than low or medium
Q. Where can free users try Hydrogen?
- Free users get Hydrogen in Antigravity, Google AI Studio's free tier (where inputs improve Google products) and via auto selection in the Gemini app