Question 1

What is Gemini 4 Hydrogen?

Accepted Answer

Google describes it as a Gemini 4 model built for speed and efficiency on everyday coding and agentic workflows. The frontier models in the same family — Gemini 4 Carbon and Gemini 4 Argon — are positioned for deep, long-horizon reasoning. The model card says Hydrogen is adapted from Carbon.

Question 2

When was it announced, and is it available now?

Accepted Answer

Google’s announcement is dated 12 November 2026 and says Hydrogen is rolling out globally that day across developer, enterprise, and consumer surfaces. We have not independently verified availability in every country or on every plan.

Question 3

What does the API cost?

Accepted Answer

On the paid Gemini Developer API tier, Google lists a flat price of $1.50 per million input tokens and $7.50 per million output tokens, including thinking tokens, across the full 1 million-token context window. Implicit cache reads are $0.075 per million tokens. A rate-limited free tier is also listed. Vertex / Agent Platform global list prices match those figures; non-global regions are listed higher at $1.65 input and $8.25 output.

Question 4

What is the model ID?

Accepted Answer

The published model ID is gemini-4-hydrogen, used in Google AI Studio, the Gemini API, and Gemini Enterprise Agent Platform.

Question 5

How large is the context window?

Accepted Answer

Input, thinking, and output share one unified 1,048,576-token window. Synchronous output is capped at 65,536 tokens per hop; longer outputs (up to 1M minus input) use background execution or continuation. That is a change from Gemini 3.8 Flash, which Google says separated a 1M input budget from a 64K output budget.

Question 6

What are the thinking levels?

Accepted Answer

Four: low, medium, high (the default), and a new xhigh tier. Google says to start with high and reserve xhigh for hard multi-step problems, because xhigh uses more tokens and latency and can overthink simpler prompts. thinking_budget and a minimal level are rejected with an HTTP 400.

Question 7

How is Hydrogen different from Carbon and Argon?

Accepted Answer

Google positions Carbon as its strongest frontier model and Argon as the efficient frontier model for advanced long-horizon work with lower token use than Carbon. Hydrogen is the fast execution model, also intended to run as a worker or verifier alongside those two. Carbon and Argon are not on the Gemini API free tier. Through 30 November 2026 they are listed at $2 / $10 per million input / output tokens, then $4 / $20 from 1 December 2026. They default to medium thinking and cap at high; they do not have xhigh.

Question 8

Can free Gemini app users get it?

Accepted Answer

Google says everyone can start using it in the Gemini app model dropdown and in Google Search, and that free app users get Hydrogen via automatic model selection when it is the appropriate model. The launch post named paid Google AI Pro and Ultra subscribers for Antigravity and for Gemini in Sheets (starting with those plans); Antigravity’s models documentation, checked 14 November 2026, now also lists Hydrogen for Free and Google AI Plus users. We have not tested each of these surfaces ourselves.

Question 9

What official benchmark numbers has Google published?

Accepted Answer

The DeepMind model card (November 2026) includes a table versus Claude Sonnet 5.5 and GPT-6.1 Terra. The announcement blog also cites an Artificial Analysis Intelligence Index score of 51 at the default high thinking level (54 at xhigh) versus 50 for Argon, and a 57% Terminal-Bench 4.1 score versus 18.2% for Gemini 3.8 Flash. Those are Google’s published figures, not tests run by this site. See the comparison page.

Question 10

Is Gemini 4 Hydrogen used in Google Search?

Accepted Answer

Yes. Google’s announcement says Hydrogen is upgrading Search for all users globally — better understanding of multi-part or typo-filled questions, more visual responses, and an upgraded query fan-out that surfaces a wider range of links — and lists Google Search among the surfaces where everyone can use it.

Question 11

How does Hydrogen compare with Claude Sonnet 5.5, Claude Opus 5.5, GPT-6.1 Sol and Gemini 3.1 Pro on price?

Accepted Answer

Official list prices per 1M input/output tokens, retrieved 13 November 2026: Hydrogen $1.50 / $7.50 (flat to 1M tokens); Claude Sonnet 5.5 $2 / $10; Claude Opus 5.5 $4 / $20; GPT-6.1 Sol $2 / $10 up to 272K-token prompts and $4 / $15 above; Gemini 3.1 Pro Preview $2 / $12 up to 200K and $4 / $18 above. Each has its own comparison page on this site with sources.

Question 12

Did this site test the model or collect reviews?

Accepted Answer

On 14 November 2026 we published original same-prompt tests: prompts written by this site, run once each on Gemini 4 Hydrogen, Gemini 3.8 Flash, and Gemma 4 31B on the API free tier. Those pages show playable outputs plus wall time, tokens, and equivalent paid-tier cost. We do not score the models, we do not treat those runs as a benchmark, and we do not invent user reviews or testimonials. The home page also collects public third-party experiments, reviews, products, and videos; those are other people’s work, and we have not verified their results. Where something is not in an official document, a linked public source, or one of our documented test pages, we mark it as not yet confirmed.

Question 13

Is Hydrogen a long-lived production model?

Accepted Answer

Google Cloud’s developer guide calls it a short-term generally available (GA) release, subject to a minimum 45-day retirement notice. Check Google’s model versions and lifecycle documentation for the current policy.

Question 14

What is the knowledge cutoff?

Accepted Answer

The model card says the knowledge cutoff is March 2026, and that in some domains the model’s knowledge may be less up to date. It also lists general limitations such as hallucinations, occasional slowness or timeouts, and higher token use at higher thinking levels.

Frequently asked questions