Google launches Gemini 4 Hydrogen at $1.50/$7.50 per million tokens with flat pricing across a 1M-token context
- Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, priced flat across the full 1M-token context window, and scores 51 on the Artificial Analysis Intelligence Index against Argon's 50 (54 at its xhigh thinking setting).
- Google positions Hydrogen as the fast execution model under Carbon and Argon, and the new Orchestrator in Google Antigravity switches between them inside one session: on Terminal-Bench 4.0 it reached near-Carbon task success while using 24% less quota and finishing 19% faster.
- Against Gemini 3.8 Flash, Hydrogen finishes tasks 2.5 times faster at its default high setting, lifts Terminal-Bench 4.1 from 18.2% to 57%, and cuts fabricated claims by more than 5x.
- Hydrogen is the first Gemini 4 model offered on every tier: AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and Antigravity for paid AI Pro and Ultra subscribers.
- For parallel work, Carbon can act as coordinator while Hydrogen runs worker and verifier agents that build and test changes at the same time across isolated workspaces.
Hacker News 의견들
With all these element names I keep waiting for Gemini 4 Oxygen.
This class of model is the one to watch. These are the workhorses once everyone stops tokenmaxxing and just wants work done.
Disagree, Opus-tier models are the workhorses. A $100 subscription gets you a ton of work done.
Their naming scheme is hard to follow, but at least they announced this one when it was actually available.
I pay for the min plan and tried Gemini again this week. It still can't tell me the current date and time. Benchmarks don't mean much if that's broken.
What prompt are you using? I asked the Gemini app for the time and it answered fine.
First thing I think when a Google model drops is whether I can call it from OpenRouter. There are seven ways to reach this thing and I don't know which one to pick.
It's in the second paragraph of the post.
Is this replacing Flash-Lite? It's serving Search answers. Pro and Flash at least told you the scale, nobody intuits that Hydrogen is smaller than Carbon.
From my use it's a Flash model through and through. Good subagent workhorse for implementation while a bigger model does the spec and review.
Same price as 3.8 Flash, no intro discount as far as I can tell.
A free default model this close to the frontier is going to change how the general public perceives AI.
It'll be the same story as the small OAI and Ant models: strong benchmarks, rough in real use. The gap between free and paid tiers is still huge.
The ordering is nonsense. Hydrogen, Argon, Carbon: not alphabetical, not atomic mass, not abundance, not electronegativity. Business Insider says Argon, Barium, Carbon was the internal alphabetical lineup and Hydrogen got tacked on because it's the lightest element.
At this point Starbucks sizes would be an improvement.
The post is good because the graphs compare wall-clock time and cost per task against Terra and Sonnet. My read: better than Sonnet outside coding, and at high reasoning smarter than Terra but a bit pricier.
I hope it keeps 3.8 Flash's conversational feel. Flash was the first non-SOTA model I'd actually reach for over the big ones, fast and wrote well.
In the app it writes poorly, probably the system prompt. Argon over the API wrote well for me.