Gemini 4 Hydrogen: Price, Benchmarks & How to Use

By Versus Tools AI Team · Published · Updated · 11 min read · AI Models & News

We may earn a commission if you buy through links on this page. Read the affiliate disclosure · How we review tools

Quick answer. Gemini 4 Hydrogen launched Nov 12, 2026. Verified price, benchmarks vs Gemini 3.8 Flash, the API model id, and where to use it.

Gemini 4 Hydrogen is Google’s fast Gemini 4 model for coding and agentic work, released Nov 12, 2026. On the paid Gemini API it costs $1.50 per million input tokens and $7.50 per million output tokens, flat across a 1 million-token context, and Google says it can generate up to 1 million output tokens. At the default high thinking level, Google reports an Artificial Analysis Intelligence Index of 51, against 50 for Gemini 4 Argon, and 54 at the new xhigh setting.

This page is based on official documentation, benchmarks and public reports. We did not run Gemini 4 Hydrogen, and we did not reproduce any leaderboard.

How this page was checked

We read Google’s Hydrogen announcement, published Nov 12, 2026 and last updated Nov 13, 2026, the Nov 10, 2026 post that made Gemini 4 Carbon generally available, and the Gemini Developer API pricing page. Prices below are the figures on that pricing page on Nov 16, 2026. Benchmark numbers are the ones Google prints. Where a rival has not published a matching number, the comparison table says “not published” rather than borrowing a figure from a roundup.

For the wider month, see the AI model release timeline and the best AI models ranking. The frontier Gemini 4 model is Gemini 4 Carbon. Anthropic’s Nov 9 release is Claude Fable 5.2. Alibaba’s flagship from the same week is covered in Qwen4-Max.

Key facts

Gemini 4 Hydrogen
ReleasedNov 12, 2026. The announcement was updated Nov 13, 2026.
MakerGoogle DeepMind. Byline: Koray Kavukcuoglu.
API model idgemini-4-hydrogen
Role in the Gemini 4 familyFast model for coding and agentic workflows. Google places it beside frontier models Gemini 4 Carbon and Gemini 4 Argon.
Context1 million tokens, one rate for the whole window.
Max outputUp to 1 million tokens, per the announcement and the pricing page.
Thinking levelsLow, medium, high (default), and xhigh.
Paid API price$1.50 / 1M input tokens, $7.50 / 1M output tokens, including thinking tokens.
CacheImplicit caching $0.075 / 1M tokens. Explicit caching is not supported.
Free tierListed on the Gemini API pricing page. Free-tier usage may be used to improve Google’s products. Paid tier: no.

What Google says is new

Hydrogen is not a replacement for Carbon. Google’s post says Carbon and Argon are for deep, long-horizon reasoning, and Hydrogen is for rapid execution: writing code, running tests, and verifying changes. It can also sit under those models as the execution engine in a multi-model workflow.

Four thinking levels are the control for that tradeoff. High is the default. Xhigh is new. Google says the ladder is meant to scale quality, latency, and cost so you wait and pay for the reasoning a task needs. The API price itself does not change by level. Output tokens, including thinking tokens, are billed at $7.50 per million, so a higher thinking level can still cost more because it can spend more tokens.

Google is also shipping Orchestrator in Google Antigravity. Instead of one model for a whole session, Orchestrator switches between models such as Carbon and Hydrogen. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon task success while using 24 percent less quota and finishing 19 percent faster. For parallel agents, the same post says Carbon can coordinate while Hydrogen runs worker and verifier agents. Those percentages are Google’s, and the post does not print Carbon’s standalone Terminal-Bench 4.0 score.

The consumer surfaces get a shorter version of the same pitch. Google says Hydrogen follows instructions more closely in the Gemini app, calls tools more often, and is what Search uses for a broader set of queries and more visual answers. In the Gemini app, free users are routed to it by automatic model selection when Google decides it fits the prompt, not by a manual pick on every plan.

Price

Checked on the Gemini Developer API pricing page on Nov 16, 2026.

Free tierPaid tier, per 1M tokens
InputFree of charge$1.50
Output, including thinking tokensFree of charge$7.50
Context cachingFree of charge$0.075, implicit only
Grounding with Google SearchNot available5,000 requests a month free, shared across Gemini 3 and newer models, then $14 per 1,000 requests, with a daily cap of 1,500
Used to improve Google’s productsYesNo

Google says every token is billed, including thinking tokens and the tool instructions added when you use grounding. The $1.50 / $7.50 rate does not step up inside the 1 million-token window. That is the contrast with older Gemini prices that changed after a context threshold. Hydrogen does not add a second tier.

Gemini 4 Carbon, on the same page, is not free. Through Nov 30, 2026 its paid rate matches the introductory Argon rate: $2 per million input tokens and $10 per million output tokens. From Dec 1, 2026 the pricing page lists $4 and $20. Implicit caching follows the same split, $0.10 then $0.20. The Carbon announcement states the same introductory window and the same post-introduction price.

Gemini 3.8 Flash remains the cheaper Google API through the end of 2026: $0.75 input and $3.75 output per million tokens until Dec 31, 2026, then $1.50 and $7.50 from Jan 1, 2027. Its caching price is $0.075 per million through that date, plus a storage price of $0.50 per million tokens per hour. Hydrogen’s pricing row does not list a cache storage fee, and it does not offer explicit caching.

A paid Google AI Pro or Ultra subscription is a separate gate. The announcement says those subscribers can use Hydrogen in Google Antigravity. Carbon’s Nov 10 post limits the Antigravity, Gemini app, and Search AI Mode rollout to Google AI Ultra.

Benchmarks Google published

All of the scores in this section come from Google’s Hydrogen post. We did not rerun them.

Against Gemini 4 Argon, Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at default high thinking, versus 50 for Argon, and 54 at xhigh. Google also says Hydrogen often finishes faster than models in its class. The post does not print the latency in seconds.

Against Gemini 3.8 Flash, Google reports three headline deltas at the default high setting: 2.5 times faster, Terminal-Bench 4.1 at 57 percent versus 18.2 percent, and more than 5 times fewer fabricated claims. The post does not print the absolute fabrication rates. It also says that in early partner testing on complex graphs, SQL, and data-heavy workflows, Hydrogen ran 2 to 2.8 times faster and used 12 to 23 percent fewer tokens than 3.8 Flash. That sentence is Google describing partner tests, not a public benchmark card.

Other scores Google attributes to Hydrogen:

  • HLE-Diamond: 58.8 percent, described as leading its class.
  • Vals Tax Agent Bench: 52.3 percent, ranked number 1.
  • Vals Legal Research Bench: 55.8 percent, tied for number 1.
  • Multimodal Knowledge Agent (MKA): 53.4 percent.
  • Agents’ Last Exam v1.1: 71.6 percent.
  • Agent Arena, which Google attributes to arena.ai: number 5.

Google also says Hydrogen cuts overreach on agentic tasks by up to 81 percent and under-service by up to 79 percent versus the prior behavior it is comparing, cuts misaligned actions and task shortcutting by 12 times across its tool-use and coding evaluations, and reaches a 94 percent honesty rate on an internal agent honesty evaluation. On the Gray Swan indirect prompt-injection benchmark, Google says Hydrogen is more than 2.3 times more resilient. False refusals on HealthBench Pro are described as under 0.4 percent. Chain-of-thought legibility is given as 99.9 percent with zero deceptive reasoning edits across more than 200,000 audited turns. Those are internal or vendor-cited safety figures. The post points to a model card for the detail, and this article does not restate anything we could not see on the announcement or the pricing page.

Gemini 4 Hydrogen vs Gemini 3.8 Flash and rivals

Terminal-Bench versions do not match across labs. A 4.0 score and a 4.1 score are different tests. “Not published” means the official page we checked did not print that cell.

Gemini 4 HydrogenGemini 3.8 FlashGemini 4 CarbonClaude Fable 5.2GPT-6.1 TerraQwen4-Max
ReleasedNov 12, 2026not publishedNov 10, 2026 (general availability)Nov 9, 2026Oct 26, 2026Nov 16, 2026
API idgemini-4-hydrogengemini-3.8-flashgemini-4-carbonclaude-fable-5-2gpt-6.1-terraqwen4-max
Input / 1M tokens$1.50$0.75 through Dec 31, 2026, then $1.50$2.00 through Nov 30, 2026, then $4.00$8. Fast mode $16$1. Cached input $0.05$1.50 in Singapore. Beijing / Global $1.413
Output / 1M tokens$7.50$3.75 through Dec 31, 2026, then $7.50$10 through Nov 30, 2026, then $20$40. Fast mode $80$5$4.50 in Singapore. Beijing / Global $4.24
Context1M tokensnot published1,048,576 combined input and output on Google’s Enterprise model page. Detail in Gemini 4 Carbon1M tokensnot published1,000,000 tokens
Max outputUp to 1M tokensnot publishedUp to 1M tokens, per the pricing-page description128K tokensnot published131,072 tokens non-thinking. 262,144 tokens for thinking plus the answer
Terminal-Bench57% on 4.1 (Google)18.2% on 4.1 (Google)not publishednot published50.3% on 4.0 at maximum reasoning (OpenAI)not published
Artificial Analysis Intelligence Index51 at high, 54 at xhigh (Google)not publishednot published65 at max, 62 at high (Artificial Analysis). See Claude Fable 5.251 at max, 42 at high (Artificial Analysis)not published

Claude Fable 5.2 prices, context, and max output are from Anthropic’s model overview and the Nov 9, 2026 platform release notes. The Artificial Analysis Intelligence Index cells for Fable 5.2 and GPT-6.1 Terra are from Artificial Analysis’s comparison page. Hydrogen’s 51 is Google’s figure for Hydrogen at high thinking. Terra’s 51 is Artificial Analysis’s figure for Terra at max. Those are different measurements. Mythos 5.2 is the same underlying model with access limited to Anthropic’s verification programs, at the same listed price. GPT-6.1 Terra’s price is from OpenAI’s launch post. Its date is the date on OpenAI’s research release index and API changelog. Qwen4-Max prices are the Singapore and Beijing / Global rates on Alibaba’s pricing pages. GPT-6 Cyber, announced Nov 5, 2026, is a separate cybersecurity model. The public launch post we read does not list a token price, so it is not in this table.

The Gemini pricing page calls Carbon Google’s most capable reasoning model and says it supports long-decode generation up to 1 million tokens. The combined window in the table is from Google Cloud’s Enterprise model page, which counts input and output together. Segment limits and the /swarm preview are in Gemini 4 Carbon.

👍 Pros

  • Flat $1.50 / $7.50 pricing across a 1 million-token window, plus a free API tier
  • Google reports a large Terminal-Bench 4.1 jump versus Gemini 3.8 Flash, 57 percent versus 18.2 percent
  • Available in AI Studio, the Gemini API, the Gemini app, Search, and Gemini for business

👎 Cons

  • Carbon is still the frontier model, and Hydrogen's scores are Google's
  • Explicit context caching is not supported
  • The free tier may be used to improve Google's products

How to use it

Google’s rollout, from the announcement:

  • Developers: the Gemini API in Google AI Studio and Android Studio. The pricing page lists a free tier and the paid rates above. The model id is gemini-4-hydrogen.
  • Google Antigravity: paid Google AI Pro and Ultra subscribers. This is where Orchestrator can switch between Carbon and Hydrogen.
  • Business: Gemini for business, through the API or the model menu, plus Antigravity, and Gemini in Google Sheets for eligible business subscriptions.
  • Gemini app: the model menu. Google says free users receive Hydrogen through automatic selection when it fits the prompt.
  • Google Search: rolling out for users generally. Carbon in the Gemini app and in Search AI Mode is the Ultra tier, per the Carbon post.
  • Google Sheets for individuals: Google AI Pro and Ultra, to start.

We are not publishing a sample request. Endpoint details and SDK shapes change, and the pricing page is the source for the model id. Confirm the id in AI Studio before you migrate a 3.8 Flash workflow.

If the work is the editor around the model rather than the model itself, Cursor is the coding tool in this site’s catalog. Hydrogen’s launch is about the model Google says is built for that kind of daily coding loop. Cursor’s own plan and model menu are separate, and this article did not re-check them on Nov 16, 2026.

See Cursor →

Where Hydrogen fits this month

Use Hydrogen when you want a Gemini 4 model that is on the free and paid API, priced in one flat band, and aimed at coding and agent loops. Use Carbon when Google’s own framing matches the job: the hardest multi-step coding, scientific reasoning, and cybersecurity work, on a paid API, at the introductory $2 / $10 rate through Nov 30, 2026. Use Gemini 3.8 Flash if the bill matters more than the Terminal-Bench jump and you are content with the lower rate that lasts through Dec 31, 2026.

Qwen4-Max, released Nov 16, 2026, is the other flagship from this week. Its Singapore list price is $1.50 input and $4.50 output per million tokens, with a much lower implicit cache-hit input rate. The two posts do not share a benchmark, so this page does not declare a winner.

The release timeline lists the dated launches around Hydrogen, including Carbon, Claude Fable 5.2, and GPT-6.1 Terra. The November 2026 ranking sorts those models by job, using the same official pages.

These are products in our catalog, not models inside the Gemini API. Commissions do not change the model comparison above.

  • Cursor for the editor, if the reason you care about Hydrogen is repository coding.
  • Fireflies for meeting notes and searchable call transcripts, which is a different job from a long-context model.
  • Writesonic for marketing drafts, if you want a writing product rather than the Gemini app.

Sources

Read on Nov 16, 2026, unless the page prints its own date.

Pricing and availability last checked Nov 16, 2026. This page contains affiliate links. A commission does not change the model facts.

Frequently asked questions

When was Gemini 4 Hydrogen released?

Google published the Gemini 4 Hydrogen announcement on Nov 12, 2026. The same post was last updated Nov 13, 2026. The model id on the Gemini API pricing page is gemini-4-hydrogen.

How much does Gemini 4 Hydrogen cost?

On the paid Gemini Developer API, checked Nov 16, 2026, input is $1.50 per million tokens and output, including thinking tokens, is $7.50 per million tokens. Google says that rate is flat across the 1 million-token context window. Implicit context caching is $0.075 per million tokens. Explicit caching is not supported. The free tier lists input and output as free of charge, and Google says free-tier prompts may be used to improve its products.

How does Gemini 4 Hydrogen compare with Gemini 3.8 Flash?

Google's launch post says Hydrogen finishes tasks 2.5 times faster at its default high setting, scores 57 percent on Terminal-Bench 4.1 against 18.2 percent for Gemini 3.8 Flash, and produces more than 5 times fewer fabricated claims. Those figures are Google's. On the API pricing page, Gemini 3.8 Flash is cheaper through Dec 31, 2026, at $0.75 per million input tokens and $3.75 per million output tokens, then moves to $1.50 and $7.50 on Jan 1, 2027.

What is the Gemini 4 Hydrogen API model name?

The Gemini Developer API pricing page lists the model id as gemini-4-hydrogen. Google says it is available in Google AI Studio and the Gemini API, including a free tier on that pricing page.

Where can I use Gemini 4 Hydrogen?

Google's post says it is rolling out in Google AI Studio, the Gemini API, Android Studio, Gemini for business, the Gemini app, and Google Search. Paid Google AI Pro and Ultra subscribers can use it in Google Antigravity. In the Gemini app, Google says free users get it through automatic model selection when it is the appropriate model for the prompt.

Is Gemini 4 Hydrogen the same as Gemini 4 Carbon?

No. Google describes Carbon and Argon as the frontier models for deep, long-horizon work, and Hydrogen as the faster model for everyday coding and agentic workflows. Carbon's paid API price, checked Nov 16, 2026, is $2 per million input tokens and $10 per million output tokens through Nov 30, 2026, then $4 and $20. Carbon is not on the free tier. Hydrogen is.