Gemini 4 Hydrogen Pricing: Free Tier, $1.50 / $7.50, and What to Move Before Carbon and Argon Double on December 1

Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens on the Gemini API's Standard tier, thinking tokens included. Batch and Flex cost half that ($0.75 / $3.75), and unlike Carbon and Argon, it has a free tier. These are Google's numbers from the Gemini API pricing page, checked Nov 13, 2026, the day after launch.

The price matters because of a date. Gemini 4 Carbon and Gemini 4 Argon go from $2 / $10 to $4 / $20 on December 1, 2026. After that, the same traffic costs 62.5% less on Hydrogen than on Carbon or Argon. Google also says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default setting, one point above Argon's 50.

Below: the full Hydrogen rate card, what the free tier does and doesn't include, monthly costs for three common workloads next to Carbon, Argon, Gemini 3.8 Flash and three Claude models, and which workloads we would test on Hydrogen before December 1.

How much does Gemini 4 Hydrogen cost per million tokens?

All prices are USD per 1 million tokens, from Google's pricing page for gemini-4-hydrogen, read Nov 13, 2026.

A few details from the same page that change the bill:

• Google's launch post says Hydrogen is "priced flat across its entire 1 million-token context window", and the pricing page lists no 200K threshold, so long prompts carry no surcharge.
• Hydrogen has one price on the page. Carbon and Argon each show a second one "starting December 1, 2026".
• Caching is implicit only. Explicit caching is not supported, so you don't pay storage, but you can't pin a cache either. Cache hits bill at $0.075, 5% of the input price.
• Thinking tokens bill as output. Google states that every token the model processes is counted, "including thinking tokens". A request at the xhigh thinking level writes more of them than one at low.
• Grounding with Google Search includes 5,000 free requests per month on paid tiers, shared across all Gemini 3 and newer models. It is not available on the free tier.

The short version: Hydrogen's Standard price is exactly 37.5% of what Carbon and Argon will cost from December 1, on both input and output. Batch keeps the same ratio.

Does Gemini 4 Hydrogen have a free tier?

Yes. On the Standard tier, input, output and context caching are all listed as "Free of charge" for free-tier projects. Carbon and Argon have no free tier at all; the pricing page says so in plain words for each.

Two limits come with it:

1. Your prompts can be used to improve Google's products. The free-tier column marks "Used to improve our products" as "Yes". On paid tiers it says No. Keep customer data and anything under an NDA off a free-tier key.
2. Search and Maps grounding are "Not available" on the free tier.

Google's pricing page describes the free tier as "Limited access to certain models" and reserves higher rate limits for paid tiers, so check your own project's limits before you plan a prototype around it. For prototyping, evals on public data and personal tools, the free tier is the cheapest way to see if Hydrogen handles your prompts. Our free AI models catalog lists the other Gemini models you can call at no cost.

What changes on December 1, and what doesn't

Nothing about Hydrogen changes on December 1. Its neighbours do.

Input / output per 1M tokens, all from Google's pricing page, Nov 13, 2026.

The Gemini 3.8 Flash row surprises people. Flash is still half Hydrogen's price until December 31, 2026. On January 1, 2027 Google's page lists Flash at $1.50 / $7.50, the same as Hydrogen. So if you're already on 3.8 Flash, decide whether Hydrogen's quality gain is worth paying double for the next seven weeks, or switch on January 1 when the price gap closes.

For teams on Carbon or Argon, the deadline is December 1.

What will Gemini 4 Hydrogen cost per month?

We reused the three workloads from our Gemini 4 Carbon pricing breakdown so the numbers line up. Every example uses list prices and assumes no cache hits. Output counts are total output per request, thinking included. Measure real output tokens in your logs before you trust any estimate, ours included.

Workload 1: support chatbot

• 1,000,000 requests per month
• 1,500 input tokens and 300 output tokens per request
• Total: 1,500M input tokens, 300M output tokens

A live chatbot can't wait for Batch, so Standard is the real comparison. Moving this workload from Carbon to Hydrogen in December saves $7,500 a month on list price.

Workload 2: RAG / document Q&A

• 100,000 questions per month
• 8,000 input tokens (retrieved chunks plus question) and 500 output tokens each
• Total: 800M input tokens, 50M output tokens

RAG is input-heavy, and Hydrogen's flat pricing across the 1M window means a long retrieved context doesn't push you into a higher tier.

Workload 3: overnight batch summarization

• 200,000 documents per month
• 20,000 input tokens and 1,000 output tokens each
• Total: 4,000M input tokens, 200M output tokens

If summaries only need to be ready by morning, Hydrogen on Batch costs 37.5% of Carbon on Batch after December 1.

Gemini 4 Hydrogen vs Carbon, Argon, 3.8 Flash and Claude on price

Same three workloads, same token counts, Standard tier, no caching. Claude prices are from Anthropic's pricing page, checked Nov 13, 2026.

Price per token is half the story. Claude Haiku 5.5 is far cheaper per token than Hydrogen, and Gemini 3.8 Flash is half its price until the new year. Google's case for Hydrogen rests on its benchmark scores (next section), so the comparison that matters is quality per dollar on your own prompts. We don't have a same-source quality score for every model in this table, so we haven't ranked them on quality.

Where you are paying Carbon, Argon or Opus 5.5 prices today for work a cheaper model could handle, Hydrogen is the first model to test. For the full menu of cost levers beyond model choice (caching, batching, context trimming), see our AI API cost optimization guide.

Is Gemini 4 Hydrogen as good as Argon?

On Google's own numbers, close. Google's launch post (Nov 12, 2026, Koray Kavukcuoglu) reports:

Artificial Analysis's own model page lists the xhigh variant at 54, with an output speed of 188 tokens per second, read Nov 13, 2026.

Google also says Hydrogen finishes tasks 2.5 times faster at its default high setting, in a passage aimed at teams moving from 3.8 Flash, and that partners saw 12% to 23% lower token usage against 3.8 Flash on complex graph, SQL and data-heavy workflows. Those are vendor and partner figures, not independent tests.

Two caveats before you route on these numbers:

• The thinking level changes the price. The 54 score is at xhigh, which writes more thinking tokens, and those bill at $7.50 per million. Test at high (the default) first and only raise it where your evals show a gap.
• An index is an average. A one-point lead over Argon on a composite index says little about your extraction schema or your agent's tool calls. Your own evals decide.

Which workloads to move to Hydrogen before December 1

We haven't run Hydrogen in a client production workload yet; it launched on November 12. This is how we would set up the test with 18 days left at the old Carbon and Argon prices.

Good first candidates for Hydrogen

• High-volume chat and support replies where Carbon or Argon was chosen for quality, not because a cheaper model failed
• RAG answers over long retrieved context, where flat pricing to 1M tokens helps
• Extraction, classification and routing steps inside an agent
• Sub-agent and tool-calling steps that run many times per task
• Overnight summarization and backfills on Batch at $0.75 / $3.75

Keep on Carbon, Argon or Claude until your evals say otherwise

• The hardest reasoning steps where your current model clearly beats a smaller one
• Workloads already tuned and evaluated on a Claude model, where switching vendors costs more engineering time than it saves
• Anything that relies on explicit caching (Hydrogen supports implicit caching only)

A routing setup usually ends up with two or three models: a cheap one for most traffic, a stronger one for the requests that need it. Google is doing a version of this itself: its Antigravity post describes an Orchestrator that pairs Hydrogen with Carbon or Argon. Google says it "reached near-Carbon task success rates while, on average, consuming 24% less quota", without stating the baseline for the quota figure.

A 5-step plan for the next 18 days

1. Pull a week of real requests per workload, with input and output token counts.
2. Re-run them on Hydrogen at the default high thinking level and score the answers against your current model.
3. Move every workload that passes. Leave the rest on Carbon, Argon or Claude.
4. Move offline jobs to Batch, whichever model they land on.
5. Compare daily spend on December 2 with the last week of November.

If you want help building the eval set and the routing rules, our fractional CAIO and AI strategy team does this work on model spend, and if the routing lives inside an agent, our AI agent development team builds it.

FAQ

How much does Gemini 4 Hydrogen cost?

$1.50 per million input tokens and $7.50 per million output tokens on the paid Standard tier, with thinking tokens billed as output. Batch and Flex cost $0.75 and $3.75. Priority costs $2.70 and $13.50. Prices from Google's Gemini API pricing page, checked Nov 13, 2026.

Is Gemini 4 Hydrogen free?

Yes, on the free tier of the Gemini API. Input, output and caching are free of charge, but free-tier content can be used to improve Google's products and Search grounding isn't available. Paid tiers don't use your data that way.

Is Gemini 4 Hydrogen cheaper than Gemini 4 Carbon?

Yes. Through November 30, Carbon costs $2 / $10, so Hydrogen is 25% cheaper. From December 1, Carbon costs $4 / $20, and Hydrogen costs 62.5% less on both input and output. Our Gemini 4 Carbon pricing guide has the full Carbon rate card.

Gemini 4 Hydrogen vs Gemini 3.8 Flash: which is cheaper?

Gemini 3.8 Flash, until December 31, 2026. It costs $0.75 / $3.75, half of Hydrogen. Google's pricing page lists Flash at $1.50 / $7.50 from January 1, 2027, the same as Hydrogen, so the price gap closes then.

Does Gemini 4 Hydrogen charge more for long prompts?

No. Google says Hydrogen is priced flat across its full 1 million-token context window, and the pricing page lists no 200K threshold. Long RAG and document prompts cost the same per token as short ones.

What is the Gemini 4 Hydrogen context window?

1 million input tokens. Google's launch post says output can reach up to 1 million tokens, while the models page lists max output as "65K / 1M", so check the output limit your SDK or tool sets by default.

Is Gemini 4 Hydrogen better than Argon?

On one index, slightly. Google reports 51 on the Artificial Analysis Intelligence Index at default high thinking versus 50 for Argon, and 54 at xhigh. That is a vendor-reported composite score, so run your own prompts before moving work.

Do thinking tokens count toward the Hydrogen price?

Yes. Google bills thinking tokens as output at $7.50 per million on Standard. Higher thinking levels such as xhigh produce more of them, so the same prompt can cost more at xhigh than at the default high.

---

Sources: Google Gemini API pricing, Google Gemini API models, Google: Introducing Gemini 4 Hydrogen, Google Antigravity: Gemini 4 Hydrogen in Antigravity, Artificial Analysis: Gemini 4 Hydrogen, Anthropic pricing. All checked Nov 13, 2026.