Gemini 4 Hydrogen Pricing: $1.50 Input, $7.50 Output per 1M Tokens

Quick answer: Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens on the paid Gemini API, and that rate stays flat across the whole 1 million-token context window. Batch and Flex requests are half price ($0.75 / $3.75), and the free tier lets you call it at no cost if Google can use your prompts to improve its products.

Last updated: November 15, 2026

Gemini 4 Hydrogen API price by tier

Google released Hydrogen on November 12, 2026 as the fast, everyday model in the Gemini 4 family. These are the official per-million-token prices from the Gemini API pricing page.

Tier Input Output (thinking included) Cached input Your data used to improve Google products?
Free Free of charge Free of charge Free of charge Yes
Standard (paid) $1.50 $7.50 $0.075 No
Batch / Flex $0.75 $3.75 $0.0375 No
Priority $2.70 $13.50 $0.135 No

A few billing details matter more than the headline rate:

  • No long-context surcharge. Google prices Hydrogen flat across its entire 1M-token window, so a prompt that fills most of the window costs the same per token as a short one.
  • Thinking tokens are billed as output. Every token the model reasons with counts at the $7.50 output rate, along with internal tool instructions and function declarations when you use grounding.
  • Caching is automatic only. Hydrogen does not support explicit caching (`cachedContents`). Implicit caching kicks in on prompts of at least 4,096 tokens and cuts cached input by 95%, to $0.075.
  • Search grounding is extra. Paid accounts get 5,000 free Google Search grounding requests a month, shared across Gemini 3 and newer models, then $14 per 1,000, capped at 1,500 requests a day. The free tier can't use grounding at all.

Hydrogen vs. 3.8 Flash, Argon and Carbon prices

Hydrogen launched into a lineup where two price changes are already scheduled. Here is the standard paid rate for each model, now and after the announced changes.

Model Price now (input / output) Scheduled change
Gemini 4 Hydrogen $1.50 / $7.50 None announced
Gemini 3.8 Flash $0.75 / $3.75 $1.50 / $7.50 from January 1, 2027
Gemini 4 Argon $2.00 / $10.00 $4.00 / $20.00 from December 1, 2026
Gemini 4 Carbon $2.00 / $10.00 $4.00 / $20.00 from December 1, 2026
Gemini 3.5 Flash-Lite $0.30 / $2.50 None announced

Two things stand out. 3.8 Flash is half Hydrogen's price only until the end of 2026; from January 1 the two cost exactly the same. And Argon and Carbon have no free tier, while Hydrogen does.

What one job costs at each date

Take a single job that sends 1 million input tokens and gets back 200,000 output tokens, thinking included, at standard rates:

Model November 2026 December 2026 January 2027
Hydrogen $3.00 $3.00 $3.00
Hydrogen (Batch) $1.50 $1.50 $1.50
3.8 Flash $1.50 $1.50 $3.00
Argon or Carbon $4.00 $8.00 $8.00
3.5 Flash-Lite $0.80 $0.80 $0.80

If you are on 3.8 Flash today, staying put saves money for about six more weeks.

After that, Hydrogen is the same price and, by Google's own testing, finishes tasks 2.5 times faster at its default setting and scores 57% on Terminal-Bench 4.1 versus 18.2% for 3.8 Flash. Google lists Hydrogen as the recommended upgrade from 3.8 Flash and older Gemini 3 and 2.5 models.

Thinking levels change your bill more than the price list

Hydrogen has four thinking levels: low, medium, high (the default) and xhigh. Because thinking tokens bill at the output rate, the level you pick decides how much you really pay. Artificial Analysis, the independent benchmark site Google cites, ran its Intelligence Index at every level:

Thinking level Intelligence Index score Output tokens used on the test Time per task
low 27 20.3M 1.4 min
medium 41 54.4M 3.8 min
high (default) 51 110M 8.3 min
xhigh 54 304M 14.7 min

Going from high to xhigh buys 3 points but produces almost three times the output tokens, which means almost three times the output bill. Artificial Analysis puts the xhigh run at $3.16 per task. Google says the same thing in plainer words: xhigh increases latency and token use and can overthink simple prompts.

A practical way to set it:

  1. Start at high, the default. It is the setting where Hydrogen edges out Argon, 51 to 50.
  2. Try medium on your own test prompts. It used about half the output tokens of high on the index.
  3. Use low for chat, routing and bulk processing where speed matters most.
  4. Keep xhigh for hard, multi-step coding or agent tasks where high falls short.

The Artificial Analysis figures are as of November 15, 2026, and the site updates them as it reruns tests.

Is Hydrogen better than Argon for the money?

At the default setting, yes on the headline index: Hydrogen scores 51 to Argon's 50 while charging $1.50 / $7.50 against Argon's $2.00 / $10.00, which doubles on December 1.

Google still positions Argon and Carbon for deep, long-horizon reasoning, and its own advice is to pair them: let Carbon plan or untangle hard problems, then hand code edits, subagent tasks and test runs to Hydrogen.

Against rivals, Google's model card compares Hydrogen with Claude Sonnet 5.5 and GPT-6.1 Terra. Hydrogen leads on 7 of the 10 benchmarks, including SWE-sweep (13.5% vs. 3.1% for Sonnet 5.5) and Agents' Last Exam v1.1 (71.6% vs. 60.4%). Sonnet 5.5 wins on three:

Benchmark Hydrogen Claude Sonnet 5.5 GPT-6.1 Terra
Terminal-Bench 4.1 57.0% 63.9% not listed
OSWorld-2.1 (computer use) 77.6% 83.9% 74.2%
Terminal-Bench-Science 0.1 51.4% 59.9% 58.3%

If your agent mostly drives a terminal or a desktop, weigh those three rows before switching. These are Google's numbers, not an independent test.

Free tier: what you give up

You can call `gemini-4-hydrogen` on the free tier of the Gemini API and in Google AI Studio for free. Three trade-offs come with it:

  • Your prompts and responses can be used to improve Google's products. On paid tiers they are not.
  • Search and Maps grounding are not available.
  • Batch and Flex are paid-only.

For personal projects and prototypes, that is a good deal. For customer data or company documents, move to the paid tier.

Moving from 3.8 Flash: changes that affect cost

Swapping the model ID from `gemini-3.8-flash` to `gemini-4-hydrogen` works, but a few defaults change:

  • Default thinking is higher. Hydrogen defaults to high, not medium, so the same prompt can use more output tokens. Test medium if your costs jump. Sending `minimal` now returns an error.
  • One shared 1M window. 3.8 Flash had separate 1M input and 64K output budgets. Hydrogen shares 1,048,576 tokens across input, thinking and output.
  • 64K per synchronous response. Responses top out at 65,536 tokens per request. For longer output, set `max_output_tokens` higher and `background` to true, up to 1,048,576 minus your input.
  • Explicit caches go away. Remove `cached_content` fields and rely on implicit caching.
  • Sampling settings are ignored. Remove `temperature`, `top_p` and `top_k`; out-of-range values return a 400 error.

Moving from Argon or Carbon needs no request changes.

Using Hydrogen without the API

If you just want Hydrogen in the Gemini app, you pay for a Google AI plan, not per token. US prices and what each plan gets:

Plan (US) Price Chat models Task models
Free $0 Auto or Flash-Lite None
Google AI Plus $4.99/month Flash-Lite, Hydrogen None
Google AI Pro $19.99/month Flash-Lite, Hydrogen, Pro, Argon Hydrogen, Argon
Google AI Ultra $99.99 or $199.99/month Flash-Lite, Hydrogen, Argon, Carbon Hydrogen, Argon, Carbon

Free users can't pick Hydrogen from the menu. With auto model selection on, most prompts go to Flash-Lite, and only prompts that need more reasoning may use Hydrogen. With auto off, every answer comes from Flash-Lite. AI Plus at $4.99 is the cheapest way to choose Hydrogen directly. The app offers Low, Medium and High thinking levels; xhigh is API-only.

Hydrogen also powers Google Search for all users globally at no charge.

In Google Antigravity, Google's coding agent app, it is rolling out to paid subscribers, along with Orchestrator, which switches between Hydrogen and Argon or Carbon inside one task. Google says Orchestrator came close to Carbon's success rate on Terminal-Bench 4.0 while using 24% less quota on average.

Sources

Prices are Google's US list prices as of November 15, 2026, and Google can change them. Benchmark results are from Google and Artificial Analysis as published on that date.