Gemini 4 Hydrogen announcement graphic from Google

Gemini 4 Hydrogen: Who Gets It, and the Two Settings That Decide

Google shipped Gemini 4 Hydrogen on 12 November and framed it as bringing Gemini 4 to everyone. Its own model help page says an account with no Google AI plan cannot select the model in a chat at all. Both statements are true, and the gap between them decides whether you ever touch it.

Key takeaways

  • Free accounts get one chat option: Auto. Google’s model table lists no selectable model for an account with no Google AI plan.
  • Task mode is the exception. The same table gives a no-plan account Hydrogen in Tasks, with no asterisk and no routing.
  • The app caps reasoning at High. The xhigh tier in the announcement is API and AI Studio only, and manual selection defaults to Low.
  • Search got the model globally, with a wider query fan-out and no published figure for how much wider.
  • On the API it is $1.50 and $7.50 per million input and output tokens, free on the Free Tier, with thinking tokens billed as output.

What Gemini 4 Hydrogen is, and what “available to everyone” leaves out

The announcement went up on 12 November, written by Koray Kavukcuoglu, SVP at Google DeepMind. It positions the model as the daily driver for agentic workflows, with Carbon held back for long-horizon reasoning.

The figures are speed figures. Google puts it at 51 on the Artificial Analysis Index at its default high thinking level, against 50 for Gemini 4 Argon, rising to 54 at xhigh. Against Gemini 3.8 Flash, the model it supersedes, Google reports 2.5 times faster completion at that default. It also claims 57 percent against 18.2 percent on Terminal-Bench 4.1, and fabricated claims cut by more than five times. Partners in early testing ran 2x to 2.8x faster on 12 to 23 percent fewer tokens.

Every one of those figures is Google’s own, with no independent replication published. Treat them as a direction of travel.

Then the access question. The announcement tells everyone to start using it in the Gemini app model drop down. Its own footnote narrows that at once. Free users get the model through auto model selection, and only when it is the appropriate model for the prompt.

Google’s model availability table is blunter. For an account with no Google AI plan, the available models in Chats column contains exactly one entry, Auto. There is nothing to pick. The page also says manual model switching, in both Chat Mode and Task Mode, is for accounts with a Google AI plan. The drop-down the announcement points at is a paid control.

Auto is not neutral either. Google describes it as selecting Hydrogen or a fast, lightweight model, along with the appropriate thinking level, while conserving your usage. It is a budget router first. And if you do pay, the page says reaching your usage limit drops your chats to that lightweight model.

Gemini 4 Hydrogen availability by Google AI plan in the Gemini app model menu

Task mode is how a free account gets Gemini 4 Hydrogen

The useful half of that table is the next column. For an account with no Google AI plan, the available models in Tasks is Hydrogen. Flat, no asterisk, no routing logic.

So free in chat means maybe, and free in Tasks means yes. If you are on the free plan and the job actually matters, run it as a Task instead of typing it into a chat box. That is one behaviour change available today at no cost.

Two limits. Tasks is a different surface from a conversation, so it is not a drop-in replacement for back-and-forth work. And Google publishes no Task allowance for the free plan, only the general note that more advanced models and higher thinking levels use more of your usage.

The app stops at High thinking, and defaults to Low

The announcement makes a lot of four thinking levels: low, medium, high as the default, and a new xhigh tier. The help page lists three for the app. Low, Medium, High. No xhigh, which puts that tier in the API and AI Studio only.

The expensive detail is the default. On the API the default is high. In the app, Google says that if you do not select a thinking level when you manually select a model, it will default to Low. Same model, materially weaker, because of a control you did not touch.

If you are on a paid plan and you pick Hydrogen by hand, set the thinking level in the same action. It costs one extra tap, and it is the difference between the model Google benchmarked and the cheapest version of it.

In Search, the query fan-out got wider with no numbers attached

The dedicated Search post is by Robby Stein, VP of Product for Google Search. He calls it the biggest model upgrade to Search yet, available globally for all users.

Google’s heading for it is “Improved query fan-out and linking”. The model’s reasoning strengthens that fan-out, connecting people to a broader range of helpful links than before. Search runs simultaneous queries across subtopics, to help people find sites they previously might not have discovered.

Google publishes no figure for how much broader, no before-and-after, and nothing addressed to site owners. The post does not mention publishers once, and never names AI Overviews. You cannot size this from Google’s post, so the only honest move is to measure it on your own properties. If you have never set up AI search visibility tracking, this is the week, and the groundwork is the same groundwork as answer engine optimization.

What Gemini 4 Hydrogen costs if you build on it

The pricing page lists it as gemini-4-hydrogen. On the paid tier it is $1.50 per million input tokens and $7.50 per million output tokens, flat across the whole 1 million-token context window. On the Free Tier it is free of charge, and context caching is $0.075, implicit only.

Three things belong in a budget. Google states that every token the model processes is counted and billed, including thinking tokens, so raising the thinking level raises the invoice. Grounding with Google Search is 5,000 free requests a month, shared across Gemini 3 and newer, then $14 per 1,000 with a daily cap of 1,500. It is not available on the Free Tier at all. And the Free Tier is marked as used to improve Google’s products, where the paid tier is not.

The comparison worth running is against what it replaces. Gemini 3.8 Flash is $0.75 and $3.75 through 31 December 2026, then $1.50 and $7.50 from 1 January. So Hydrogen costs double its predecessor today, and that gap closes in January because the older model doubles, not because Hydrogen gets cheaper.

Gemini 4 Hydrogen API pricing per million input and output tokens

Your checklist for this week

  1. Check which model answered you. Open the response menu and select See response details, which names the model used.
  2. On the free plan, run it as a Task. Chats give you Auto; Tasks give you Hydrogen outright.
  3. Set the thinking level yourself. Manual selection defaults to Low, not high.
  4. Plan for the drop. On a paid plan, hitting your usage limit moves your chats to the lightweight model.
  5. Budget thinking tokens as output. At $7.50 per million they are billed like any other output token.
  6. Re-read the table before you quote it. Google says it is subject to change without notice, and it changed this week.

The app’s model list now reads Hydrogen, Neon and Carbon, with Neon appearing only in the Pro and Ultra rows, while the 12 November announcement still discusses Argon. Google added Gemini 4 Neon after Hydrogen shipped and the help page moved with it. Anything you write down about plan tiers has a shelf life of days.

Questions this raises

Can I use Gemini 4 Hydrogen on the free plan?

In Tasks, yes, directly. In chats, only when Auto routes you to it, because Auto is the sole entry in the Chats column for an account with no Google AI plan. You cannot choose it, and you cannot tell in advance which prompts will reach it.

Is the xhigh thinking level available in the Gemini app?

No. The app’s thinking levels are Low, Medium and High. The xhigh tier described in the announcement sits in the Gemini API and AI Studio, so an app subscriber cannot reach the 54 index score Google quotes for it.

Does the Search upgrade mean more traffic for my site?

There is no way to know from what Google has published. The Search post says the fan-out connects people to a broader range of links and may surface sites they had not discovered. It gives no figure, no comparison and no guidance for site owners. Measure your own referrals rather than modelling it.

The release is real and the speed gains are plausible, but nothing about it is automatic. On the free plan your lever is Task mode. On a paid plan it is the thinking level, which starts at Low whenever you choose a model by hand. On the API it is remembering that reasoning is billed as output.

The Search change is the one to watch rather than act on, because Google has published no number you could plan against. Check your referral data in three weeks against the three weeks before the twelfth.

This is the kind of thing we work on with clients at Social Lady. If you want a second opinion on which AI tools your team should actually be paying for, get in touch.

Share on Social Media

Free Consultation

Contact Us