Gemini 4 Hydrogen: price, free tier, benchmarks and my test vs Carbon

👁 111 reads

In short: Gemini 4 Hydrogen is the cheapest and fastest model in the Gemini 4 family. Since 12 November 2026 it works in the Gemini app (free users get it automatically when Google decides it fits the prompt), in Google Search and in the Gemini API, where it has a free tier. On the paid tier it costs 1.50 USD per million input tokens and 7.50 USD per million output tokens. In my test of 23 tasks it got everything right.

On the evening of 12 November 2026 Google added a third model to the Gemini 4 family. After Argon (30 September) and Carbon (available to everyone since 10 November) comes Gemini 4 Hydrogen: a model for fast everyday work, coding and agents. This time Google did not start with a small group of paying users. Hydrogen goes straight to the Gemini app, to Google Search and to the Gemini API, including the free tier. I ran it the same evening through the same 23-task set I used for Argon and Carbon.

The most capable model of the family is covered in my Gemini 4 Carbon guide.

What Google announced

Google calls Hydrogen its model “for everyday agentic workflows”. Argon and Carbon are meant for long, hard reasoning; Hydrogen is meant to execute tasks quickly, on its own or as a “high-efficiency execution engine” next to Carbon. In Antigravity, the Orchestrator mode switches between Carbon and Hydrogen within one session.

• API model ID: gemini-4-hydrogen, stable (GA) since 12.11.2026. On OpenRouter: google/gemini-4-hydrogen.
• Context: 1,048,576 tokens, shared by input and output. Output up to 65,536 tokens by default, up to 1,048,576 with long decode.
• Input: text, images, video, audio and PDF. Output: text only (no image generation, no Live API).
• Thinking: four levels: low, medium, high (default) and the new xhigh.
• Tools: function calling, code execution, computer use, grounding with Google Search and Google Maps, URL context, structured outputs.

Where it is available

In the Google AI plans, Hydrogen is listed in Google AI Plus, Pro and Ultra. Argon starts at Pro, and Carbon is only in Ultra. Google describes the rollout as global and gives no country list, so the app may show Hydrogen gradually.

API pricing

Prices per million tokens, output includes thinking tokens. Gemini API pricing page, checked 12.11.2026.

One price covers the whole context window, with no surcharge for long prompts. Priority mode costs 2.70 and 13.50 USD. The biggest difference for hobbyists is the free tier: Carbon and Argon have none in the API, Hydrogen does. Google’s pricing page says free-tier data may be used to improve its products, but the Gemini API terms state that in the European Economic Area, Switzerland and the United Kingdom the paid-service data terms apply to all services, including the free quota and AI Studio.

Benchmarks: Google’s numbers and independent ones

Google says Hydrogen scores 51 in the Artificial Analysis Intelligence Index at high (Argon: 50) and 54 at xhigh. On Terminal-Bench 4.1 it reports 57% versus 18.2% for Gemini 3.8 Flash, 58.8% on HLE-Diamond, and tasks finished 2.5 times faster than 3.8 Flash at the high setting. It also claims more than 5 times fewer fabricated claims than 3.8 Flash. These are Google’s numbers.

Artificial Analysis measured Hydrogen on launch day at four thinking levels:

Artificial Analysis, API v2, checked 12.11.2026, 18:04 UTC. Index cost at AA prices (for Argon and Carbon AA already uses the 4/20 USD price).

Google’s claim holds: at high, Hydrogen matches Argon at a fraction of the cost. The thinking level matters a lot: there are 27 points between low and xhigh. AA measures 198 tokens per second on the model page, but also high verbosity: at xhigh Hydrogen used 304 million output tokens on the index, against a median of 72.8 million. On the Arena text leaderboard it has 1490 points (15th place, preliminary), clearly behind Carbon (1526) and Argon (1522).

My test: 23 practical tasks

On launch evening I ran Hydrogen through the same set I used for Argon in October and Carbon on 10 November: 3 agent tasks with tools, 4 coding tasks checked by unit tests, proofreading, a summary and a formal letter in Polish, 11 reasoning tasks and a 158,000-character document with questions. All through OpenRouter, thinking level low, scored automatically.

Hydrogen solved all 23 tasks. The bill surprised me: despite the lower price per token I paid about the same as for Argon, because Hydrogen produced twice as many tokens (13,989 of them thinking). It was also slower than Argon. This was measured on launch evening under heavy traffic, so I may repeat the timing, but the verbosity also shows in AA’s data. Practical takeaway: a low price per token does not automatically mean a cheap task, and this set is now too easy to separate the Gemini 4 models.

Which Gemini 4 for what

• Hydrogen: everyday questions, quick tasks in the app, agents and code when cost matters; the only one with a free API tier. For harder tasks set it to high or xhigh.
• Argon: still a good general model in Google AI Pro, but in the API it costs more and at high scores like Hydrogen.
• Carbon: the hardest tasks, long reasoning, code audits. Highest score on AA and Arena, but only in Ultra and the paid API.

Kinetyra