A free-tier Gemini one point behind Sonnet on Artificial Analysis
Gemini 4 Hydrogen is free in AI Studio and hallucinated just 8% on one test.
Gemini Hydrogen found and fixed a bug that showed yen and won prices 100 times too small, in a full-stack app, in three to four minutes. It is Google's fast, cheap everyday Gemini 4 model, and you can try it free today.
The deeper look
Artificial Analysis puts it one point behind Claude Sonnet 5.5 on its index. The same firm measured an 8% hallucination rate on AA-Omniscience, against 43% for Sonnet and 52% for GPT Sol. Lower means it more often says it does not know instead of guessing.
Arena's agent board has it fifth on debut, the fastest of the top ten at a 5.3-minute median task. Google's own chart claims it matches Sonnet's best score in under half the time. That one is the maker's word.
The hands-on evidence is one reviewer, Fahd Mirza. He had it run parallel sub-agents on a loan application, and it caught the planted red flags and wrote a grounded memo declining the loan. He called it a workhorse that could stand in for Sonnet on agent work.
It is for builders who run many agent calls and want the bill to shrink, and for anyone who wants a capable free model. Skip it if your work is science-heavy reasoning or driving a computer.
How to start: open Google AI Studio and pick Gemini 4 Hydrogen, or select it in GitHub Copilot.
$ curl https://openrouter.ai/api/v1/chat/completions -H "Authorization: Bearer $OPENROUTER_API_KEY" -H "Content-Type: application/json" -d '{"model": "google/gemini-4-hydrogen", "messages": [{"role": "user", "content": "Hello"}]}'