Gemini 4 Hydrogen API Pricing
Gemini 4 Hydrogen is the fast, everyday model of Google's Gemini 4 line, generally available on the Gemini API since November 12, 2026. It costs $1.50/M input, $7.50/M output and $0.075/M cached input, one rate across the whole 1M-token window, with a free tier and Batch and Flex at half price. Google names it the upgrade from Gemini 3.8 Flash, which costs half as much until its own rate doubles to the same card on January 1, 2027. Checked by hand against Google's Gemini API pricing page.
Gemini 4 Hydrogen API cost calculator.
Live calculator on Google's paid-tier rate, which has no end date and no long-context step. Hydrogen defaults to the high thinking level; dropping to medium or low cuts the output share, and xhigh raises it.
Gemini 4 Hydrogen cost on real workloads.
Codebase migration step
Find-and-patch vulnerability run
Financial research brief
Single-pass module rewrite
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Gemini 4 Hydrogen Current | $1.50 cache $0.075 | $7.50 | $0.905 agentic 92/8 | 1M | Fast, everyday Gemini 4 with a free tier |
| Gemini 3.8 Flash | $0.75 cache $0.075 | $3.75 | $0.481 −47% · promo | 1M | Half the rate until January 1, 2027 |
| Gemini 4 Neon | $3.00 cache $0.15 | $15.00 | $1.81 +2.0× | 1M | The next Gemini 4 tier up, for complex work |
| Gemini 4 Carbon | $2.00 cache $0.10 | $10.00 | $1.21 +33% · promo | 1M | Google's most capable model |
| Gemini 3.5 Flash-Lite | $0.30 cache $0.03 | $2.50 | $0.272 −3.3× | 1M | High-volume, simpler tasks |
| Claude Sonnet 5.5 | $2.00 cache $0.10 | $10.00 | $1.29 +42% | 1M | Anthropic's mid tier for agent loops |
| GPT-6.1 Terra | $1.00 cache $0.05 | $5.00 | $0.645 −29% | 1.05M | OpenAI's fast, efficient GPT-6 tier |
| DeepSeek V4.1 Flash | $0.30 cache $0.006 | $1.20 | $0.15 −6.0× | 1M | Budget reasoning and coding |
Compare siblings
AI Cost Bench is our own test: every model runs the same 82 work tasks through its maker's API and a program checks each answer. Gemini 4 Hydrogen was billed $4.09 for the whole basket at its list price. It passed all 82 tasks, so the bill is also its cost of a clean result. That is 20 tasks done right per dollar, #16 of 17 models on the index; Hunyuan Hy4 Flash is first at 343.
| Kind of work | Tasks passed | Bill | Share of the bill |
|---|---|---|---|
| Fix the bug | 6/6 | $1.18 | 29% |
| Invoice to JSON | 8/8 | $0.345 | 8% |
| Support queue triage | 40/40 | $0.881 | 22% |
| Answer from documentation | 12/12 | $0.848 | 21% |
| Back-office agent with tools | 8/8 | $0.380 | 9% |
| Data analyst | 8/8 | $0.462 | 11% |
One run, on November 14, 2026, at the model's default reasoning effort (medium): we set none. A gap of 4 tasks or less between two models is within run-to-run noise. The bill is the tokens the API reported for the run at today's list price; the clean result adds 10 cents for every failed task, and the index is the 82 tasks divided by it.
Gemini 4 Hydrogen has published scores on 4 benchmarks we track; its best placing is #4 of 6 on OSWorld 2. All of them are the lab's own reports; each row links to the page that published it.
| Benchmark | Score | Rank | Leader | Source |
|---|---|---|---|---|
| OSWorld 2 Agents and computer use · 2.1, offline subset | 77.6% offline subset | #4 of 6 | Claude Fable 5.2 · 89.3% | Vendor-reported Google DeepMind |
| Terminal-Bench-Science AI research | 51.4% as reported | #10 of 14 | Claude Mythos 5.2 · 75.3% | Vendor-reported Google DeepMind |
| Agents' Last Exam Agents and computer use · ALE v1.1 score | 71.6% v1.1, score | #3 of 3 | Gemini 4 Carbon · 77.4% | Vendor-reported Google DeepMind |
| Chartography Vision and documents · Tool use not stated | 66.6% tool use not stated | #4 of 4 | Gemini 4 Neon · 73.5% | Vendor-reported Google DeepMind |
Gemini 4 Hydrogen API pricing FAQ.
What Gemini 4 Hydrogen costs, whether it pays to move from Gemini 3.8 Flash now, and how it sits next to Carbon, Argon and the other fast tiers.
Q · 01 How much does Gemini 4 Hydrogen cost? +
$1.50/M input, $7.50/M output (thinking tokens included) and $0.075/M cached input on the paid tier, flat across the whole 1M-token window with no end date. Under this site's 92/8 agentic blend at an 82% cache-hit rate that is $0.905/M effective.Q · 02 Is Gemini 4 Hydrogen free to use? +
Q · 03 Should I move from Gemini 3.8 Flash to Hydrogen now? +
$0.75/M input and $3.75/M output through December 31, 2026, half of Hydrogen's rate. On January 1, 2027 Flash rises to $1.50 and $7.50, the same card as Hydrogen, and from then the choice is about quality and speed only. Google also says Hydrogen defaults to the high thinking level where 3.8 Flash defaulted to medium, so set the level before comparing bills.Q · 04 Hydrogen, Argon or Carbon? +
$1.50/M input against $2.00/M for Argon and Carbon through November 30, 2026, and the gap widens when their rates double on December 1. Google positions Carbon and Argon for deep, long-horizon reasoning and Hydrogen for fast execution, and suggests pairing them: Carbon plans, Hydrogen writes code and runs tests.Q · 05 What is the context window? +
1,048,576 tokens shared by input, thinking and output. A synchronous call returns up to 65,536 output tokens; with long decode and background execution one response can run up to 1M tokens minus the input. All of it bills at the output rate, and there is no higher rate for long prompts.Q · 06 Are there Batch, Flex or Priority rates? +
$0.75/M input and $3.75/M output. Priority is 1.8x: $2.70/M input and $13.50/M output. Caching is implicit only, with no storage charge, for prompts of at least 4,096 tokens.Q · 07 How does Hydrogen compare on price with Claude Sonnet 5.5 and GPT-6.1 Terra? +
$0.905/M, against $1.29/M for Claude Sonnet 5.5 and $0.645/M for GPT-6.1 Terra, the two models Google compares it with on its model card.