Google’s fast Gemini 4 model. It replaces Gemini 3.8 Flash and lists 25% below Carbon and Argon.
Edges Argon, at three times the speed.
Intelligence Index
- Carbon high60
- Hydrogen xhigh54
- Hydrogen high51
- Argon high50
- Hydrogen medium41
- Hydrogen low27
0–70 · higher is better
Output speed
- Hydrogen high236
- Hydrogen xhigh198
- Carbon high76
- Argon high64
0–250 tokens per second
View chart data
| Model and effort | Index | Tokens per second |
|---|---|---|
| Carbon high | 60 | 76 |
| Hydrogen xhigh | 54 | 198 |
| Hydrogen high | 51 | 236 |
| Argon high | 50 | 64 |
| Hydrogen medium | 41 | Not charted |
| Hydrogen low | 27 | Not charted |
Against Sonnet 5.5 and GPT-6.1 Terra.
Hydrogen leads six of the nine rows Google publishes. Sonnet 5.5 stays ahead on terminal coding and computer use.
Agents' Last Exam v1.1
- Hydrogen71.6%
- Sonnet 5.560.4%
- GPT-6.1 Terra69.8%
AutomationBench-AA
- Hydrogen76.1%
- Sonnet 5.571.8%
- GPT-6.1 Terra60.2%
Terminal-Bench 4.1
- Hydrogen57.0%
- Sonnet 5.563.9%
- GPT-6.1 Terra—Not reported
OSWorld-2.1 · offline subset
- Hydrogen77.6%
- Sonnet 5.583.9%
- GPT-6.1 Terra74.2%
View chart data
| Benchmark | Hydrogen | Sonnet 5.5 | GPT-6.1 Terra |
|---|---|---|---|
| Agents' Last Exam v1.1 | 71.6% | 60.4% | 69.8% |
| AutomationBench-AA | 76.1% | 71.8% | 60.2% |
| Terminal-Bench 4.1 | 57% | 63.9% | Not reported |
| OSWorld-2.1 · offline subset | 77.6% | 83.9% | 74.2% |
Against Gemini 3.8 Flash, Google reports tasks finishing 2.5 times faster and Terminal-Bench 4.1 rising from 18.2% to 57%.
Pick an effort, copy the request.
High is the default. Xhigh is new with Hydrogen. Set MINIROUTER_KEY to your API key.
Intelligence Index51
First answer token15.0 sInside the 60-second limit for a streamed reply to start.
curl https://api.minirouter.sh/v1/chat/completions \
-H "Authorization: Bearer $MINIROUTER_KEY" \
-H 'Content-Type: application/json' \
--data '{
"model": "google/gemini-4-hydrogen",
"reasoning_effort": "high",
"max_tokens": 32768,
"messages": [
{
"role": "user",
"content": "Review this diff for bugs and missing tests."
}
]
}'Index and wait times: Artificial Analysis, measured on Google’s API and read 12 November 2026. The line marks 60 seconds.
Before you switch.
max_tokenscounts thinking, and thinking cannot be turned off.- Google ignores
temperature,top_pandtop_k. - A streamed reply must start within 60 seconds. Xhigh takes about 41.
- Google lists one rate across the whole 1M-token window.