AI News · 5 min read
Gemini 4 Hydrogen is out: $1.50 in, $7.50 out
On November 12, 2026, Google launched Gemini 4 Hydrogen at a flat $1.50 and $7.50 per million tokens. It is the fast model for everyday work, and it is live in Search.
What Google shipped on November 12
On November 12, 2026, Koray Kavukcuoglu introduced Gemini 4 Hydrogen on Google's blog ↗. Hydrogen is the fast model for everyday coding and agentic work. Gemini 4 Carbon and Gemini 4 Argon stay the frontier models for long-horizon reasoning.
The picture above is ours. Google's hero is a wordmark.
9to5Google ↗ (Abner Li, November 12) says Google shipped Hydrogen instead of keeping the Flash name. Google's post calls it a "major standalone upgrade for teams migrating from Gemini 3.8 Flash." Flash is not fully gone from the app menu: the same 9to5Google story still lists 3.5 Flash-Lite under "Quick answers."
Where you can call it today
Google says Hydrogen is rolling out globally, for every tier of Google AI Studio and the Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro and Ultra in Google Antigravity ↗.
| Surface | Who gets Hydrogen |
|---|---|
| Gemini API, AI Studio, Android Studio | All tiers |
| Gemini app | Model menu. Free accounts through auto selection |
| Google Search | All users, globally |
| Google Antigravity | Paid Google AI Pro and Ultra |
| Gemini in Google Sheets | AI Pro, Ultra, and eligible business plans |
The app footnote is the limit that matters. Free users get Hydrogen "via auto model selection when it's the appropriate model for the prompt." They do not pin it. 9to5Google's reading of the picker puts Hydrogen on Everyday help, 3.5 Flash-Lite on Quick answers, Argon on Code and analysis, Carbon on Most complex work, and 3.1 Pro on Detailed research for AI Pro.
The price and the scores Google published
Every figure in this table is from Google's November 12 post. We did not run these tests.
| Item | Google's figure |
|---|---|
| Input | $1.50 per 1M tokens, flat across the 1M window |
| Output | $7.50 per 1M tokens, flat across the 1M window |
| Context | 1M input, up to 1M output |
| Artificial Analysis, default high | 51, against Argon at 50 |
| Artificial Analysis, xhigh | 54 |
| Terminal-Bench 4.1 vs 3.8 Flash | 57% vs 18.2% |
| Speed vs 3.8 Flash, default high | 2.5 times faster |
| Fabricated claims vs 3.8 Flash | More than 5 times fewer |
Thinking levels are low, medium, high (the default), and a new xhigh tier. The developer post ↗ ties that range to tighter instruction following and fewer fabricated claims.
What changes in Search and in Antigravity
Robby Stein, VP of Product for Search, called this ↗ "the biggest model upgrade to Search yet." Hydrogen is in Search globally for all users. Google lists three changes: better intent on multi-part questions and typos, a wider query fan-out that returns more links, and more visual responses, including images, tables, and custom tools. Stein also says the upgrade covers every supported AI Mode language.
In Antigravity, Orchestrator pairs Hydrogen with Carbon or Argon in one session. The Antigravity post ↗ says that on Terminal-Bench 4.1, Orchestrator reached near-Carbon success with 24% less quota and 19% faster completion, on average. Google's model post puts those same 24% and 19% figures on Terminal-Bench 4.0. The posts do not explain the bench-version difference. Treat them as two Google claims, not one confirmed score.
Who gets which Orchestrator, from the Antigravity post: Orchestrator (Carbon) and Orchestrator (Argon) for Google AI Ultra, and Orchestrator (Argon) for paid Google AI Pro. Enterprises get both Hydrogen and Orchestrator in Gemini for business.
What people are saying
"I'm very surprised by how fast and cheap it is. I'm very impressed by it." Tony (@EnvolDev) ↗ also posted his own run: Hydrogen at $4.27 in 1h 23m 40s, Carbon at $8.83 in 2h 39m 10s. That is his test, not ours.
"Gemini 4 Hydrogen from @GoogleAI is now available on AI Gateway." — Vercel Developers ↗
"Among closed-weight models, Gemini 4 Hydrogen is Pareto-optimal for tasks requiring a balance of speed, price, and intelligence." — Parallel (@p0) ↗, on Parallel's Search Capabilities leaderboard
"It’s really very fast, and its top-notch performance at this speed is redefining the game." — Pranav Reddy ↗
Reddit was not reachable for this story.
Our take
Developers leaving 3.8 Flash should run Hydrogen on a real repo this week. The API price is flat at $1.50 in and $7.50 out per million tokens across the full 1 million-token window. Google says default-high Hydrogen is 2.5 times faster than 3.8 Flash and scores 57% on Terminal-Bench 4.1 against 18.2%. Those numbers are Google's. Tony's single run was cheaper and faster than his Carbon run. One session is not a price rule. Judge the bill per finished task.
Do not retire Carbon. Google still points Carbon and Argon at the long jobs. A 51 on Artificial Analysis, one point over Argon, is a fast model landing on one frontier index. It is not a replacement for the model Google sells for the hardest coding and research. Carbon's introductory $2 and $10 price still doubles on December 1. Details are in our Carbon post.
In Antigravity, try Orchestrator if you already pay. Ultra gets Orchestrator (Carbon) and Orchestrator (Argon). Paid Pro gets Orchestrator (Argon). Use it for everyday feature work. Leave an all-Carbon swarm for the huge multi-worktree jobs.
Search publishers should watch the fan-out, not the API price. The model is live for every Search user, with more links and more visual answers. If AI Mode and AI Overviews send you traffic, this is the change. Plan comparisons are in ChatGPT vs Claude vs Gemini. Coding-tool context is in the best AI for coding guide and the coding assistant benchmark.
Free Gemini users do not pick Hydrogen by hand. Auto selection serves it when Google decides the prompt fits. 9to5Google still has 3.5 Flash-Lite on quick answers. A pinned model is a paid menu.