Google's fast Gemini beat its flagship on the index, and it costs less than half
Hi there,
Google just launched a fast model that beat its own flagship on the scoreboard, and it costs less than half the price.
Gemini 4 Hydrogen arrived on November 12, the quick one in the Gemini 4 family, built for everyday coding and agent work. Google says it scores 51 on the Artificial Analysis Intelligence Index at its normal setting, against 50 for Gemini 4 Argon, the flagship. Turn the thinking up to the new xhigh level and it reaches 54.
| Model | Price per million tokens | Intelligence Index | Built for |
|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 in / $7.50 out | 51 (54 at xhigh) | Everyday coding and agent work |
| Gemini 4 Carbon | $2 / $10, rising to $4 / $20 on December 1 | 59 | Deep reasoning and security work |
| Gemini 4 Argon | $2 / $10, rising to $4 / $20 on December 1 | 50 | Long, hard reasoning |
Carbon's 59 is the earlier v5.0 reading. Hydrogen and Argon are the current readings Google cites.
The fast model is the real product now. The flagship is for the hard five percent.
That is my position, and the price table is why I hold it. Hydrogen is not a small side model. It runs flat at $1.50 in and $7.50 out per million tokens across its full million-token window, while Carbon and Argon double in price on December 1.
A model that is cheaper, faster and one point ahead of the flagship on the index is not a budget option. It is the default, and the flagship is the specialist you call in when the work is truly hard.
The part that matters is not the score
One point on an index is a small win. The big win is where Google put this model. Hydrogen is in Search for all users, in the Gemini app, in AI Studio and the API on free and paid tiers, and in GitHub Copilot since November 13.
Copilot will retire Gemini 3.8 Flash on December 14 and points users at Hydrogen as the replacement.
That is the opposite of what happened with the last big launch I wrote about. Yesterday I wrote about Qwen 4, a flagship preview you can only rent for two weeks through a subscription, with no price and no benchmarks.
Hydrogen is the mirror image. No waiting room, no subscription gate, no expiry date. Google put its newest Gemini in the search box everyone already uses.
The speed numbers back it up, at least on Google's own tests. Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash, and on Terminal-Bench 4.1 it scores 57 percent against 18.2 percent for Flash. Google also says it cuts made-up claims by more than five times, which matters more than speed when a model touches real code.
There is one more piece. Google also launched Orchestrator in Antigravity, which switches between Carbon and Hydrogen inside one session. On Terminal-Bench 4.0 it reached near-Carbon results while using 24 percent less quota and finishing 19 percent faster. Carbon plans the work, Hydrogen does it, and the user never picks a model at all.
So what is a flagship for, then? The five percent of jobs where depth beats speed. Everything else now belongs to the fast model, and Google just made the fast model the one everyone already has.
The catch is in the small print
None of this makes Hydrogen perfect, and the numbers need a hard look. First, the 54 at xhigh comes with a real cost. Independent testing by Artificial Analysis found the top setting nearly doubles the output tokens, so the headline score is a paid upgrade, not a free win.
If you run everything at xhigh, the cheap model stops being cheap.
Second, every headline number here is Google's own. The speed, the 57 percent, the Orchestrator savings — all of it comes from Google's blog and Google's tests.
The independent check so far is one index reading, and on the BenchLeader index Hydrogen sits at 63.4 in 15th place, well behind Claude Fable 5.2 at 72.6. Fast is not the same as best.
Third, Hydrogen is "adapted from Carbon", in Google's own words. It is a tuned sibling of the bigger model, not a new brain. That is fine for everyday work, but the ceiling here is Carbon's ceiling, reached faster and cheaper.
Still, the direction is clear. Google spent the last month launching bigger flagships, and now it has put a cheaper, faster one in front of everyone, in Search, in the app, in the free tier. The race is no longer about who has the smartest model. It is about who puts a good enough model where people already are.
If you use Gemini or Copilot, you will meet Hydrogen this week. Try it on a real job, not a test prompt. Let it write the code, run the tests, fix the breakage, and tell me what it got wrong. I read every comment, and the bug reports teach me more than the praise.