Gemini 4 Hydrogen: Argon-class score at $1.50/$7.50

Two days ago we told you to route only the hard tasks to Gemini 4 Carbon. Today Google shipped the model for everything else. Gemini 4 Hydrogen is live in the Gemini API at $1.50 per million input tokens and $7.50 per million output. Google says it slightly beats Gemini 4 Argon on a standard intelligence index. In December, Argon will cost $4/$20. If Google's numbers hold on your work, your default model just got a lot cheaper.

What actually happened

Per Google's announcement, dated November 12, and the Gemini API pricing page:

• Model ID: gemini-4-hydrogen. It is in AI Studio, the Gemini API, Android Studio, Gemini for business, the Gemini app, and Google Search.
• Price: $1.50 input / $7.50 output, flat across the full 1 million-token context. Batch and Flex are half that. Priority is $2.70 / $13.50.
• Free tier: yes, unlike Carbon and Argon. The pricing page says free-tier content is used to improve Google's products. Keep customer data on the paid tier.
• Thinking levels: low, medium, high (the default), and a new xhigh.
• Google's benchmark claims: 51 on the Artificial Analysis Intelligence Index at high, against Argon's 50. 54 at xhigh. 57% on Terminal-Bench 4.1, against 18.2% for Gemini 3.8 Flash, and 2.5x faster than Flash.

The Agent Platform docs add three details that matter in production. Hydrogen is a "short-term" GA release with a minimum 45-day retirement notice. It ignores temperature, top_p and top_k. And caching is implicit only until January 12, 2027.

Why Gemini 4 Hydrogen matters for your business

The cheap tier stops being cheap on January 1. Gemini 3.8 Flash costs $0.75/$3.75 through December 31, then $1.50/$7.50. That is Hydrogen's price. From January, the only question is which model does your job better per dollar. Test that now.

Every benchmark here is Google's own. The model card says Hydrogen refuses more harmless prompts than 3.8 Flash, especially with images. It also says the model can burn more tokens at higher thinking levels. Run your own eval set at high and at medium, and log cost per finished task.

Check your config. If your pipeline sets temperature 0 to get repeatable output, Hydrogen ignores it. Test for drift before you switch. And a 45-day notice is short, so keep a second model wired in as a fallback.

Sources: Google: Introducing Gemini 4 Hydrogen, Gemini API pricing, Google Cloud: Gemini 4 Hydrogen docs, Google DeepMind model card.