AI model · Google

Gemini Hydrogen, Explained: Specs, Pricing, Benchmarks

Gemini Hydrogen is Google's fast, efficient Gemini 4 model for everyday coding and agent work, released November 12, 2026 at $1.50 per million input tokens. Synexa does not serve it; use Synexa when your app also needs images, video, audio or 3D from one API key.

  • 1M-token context window
  • $1.50 in / $7.50 out per MTok
  • Synexa: image, video, audio, 3D APIs
Get Started
Nov 12, 2026
Release date, rolling out globally
1M
Context window in tokens, priced flat across it
$1.50 / $7.50
Input / output price per million tokens
4
Thinking levels: low, medium, high, xhigh

What is Gemini Hydrogen?

Gemini Hydrogen, full name Gemini 4 Hydrogen, is a large language model from Google, announced on November 12, 2026. Google describes it as the model that brings Gemini 4 to everyone, built for speed and efficiency in everyday coding and agentic workflows. It reads text, images, audio, video and PDFs and writes text.

It is the fast member of the Gemini 4 family. Gemini 4 Carbon and Gemini 4 Argon are Google's frontier models for deep, long-horizon reasoning; Gemini Hydrogen is built for rapid execution, either on its own or as the worker that carries out a plan a frontier model wrote. Google's model card says it is adapted from Gemini 4 Carbon.

Before launch, press reports described Hydrogen as an internal "fast" Gemini 4 Flash model. Google's launch posts do not use a Flash name, but they pitch Gemini Hydrogen as the standalone upgrade for teams moving off Gemini 3.8 Flash, which Google's docs now call its previous-generation Flash model.

Gemini Hydrogen specs: context, output and model ID

Published figures from Google's model list and model card, as of November 2026.

Maker
Google (Google DeepMind)
Released
November 12, 2026
API model ID
gemini-4-hydrogen
Input
Text, images, audio, video and PDF
Output
Text
Context window
1M tokens, shared between input and output
Max output
65K tokens per normal response; up to 1M with background long decode
Thinking levels
low, medium, high (default), xhigh
Knowledge cutoff
March 2026
Based on
Adapted from Gemini 4 Carbon

How much does Gemini Hydrogen cost?

On the paid tier of the Gemini API, the model costs $1.50 per million input tokens and $7.50 per million output tokens, flat across the whole 1M context. Thinking tokens bill as output. Standard rates as of November 2026.

Input, per MTok

Gemini Hydrogen
$1.50
Gemini 3.8 Flash
$0.75 to Dec 31, 2026, then $1.50
Gemini 4 Carbon / Argon
$2.00 to Nov 30, 2026, then $4.00

Output, per MTok

Gemini Hydrogen
$7.50
Gemini 3.8 Flash
$3.75, then $7.50
Gemini 4 Carbon / Argon
$10.00, then $20.00

Context caching, per MTok

Gemini Hydrogen
$0.075 (implicit only)
Gemini 3.8 Flash
$0.075, then $0.15
Gemini 4 Carbon / Argon
$0.10, then $0.20

Free tier

Gemini Hydrogen
Yes
Gemini 3.8 Flash
Yes
Gemini 4 Carbon / Argon
Not available

Example: 100K in + 10K out

Gemini Hydrogen
$0.225
Gemini 3.8 Flash
$0.1125, then $0.225
Gemini 4 Carbon / Argon
$0.30, then $0.60
RateGemini HydrogenGemini 3.8 FlashGemini 4 Carbon / Argon
Input, per MTok$1.50$0.75 to Dec 31, 2026, then $1.50$2.00 to Nov 30, 2026, then $4.00
Output, per MTok$7.50$3.75, then $7.50$10.00, then $20.00
Context caching, per MTok$0.075 (implicit only)$0.075, then $0.15$0.10, then $0.20
Free tierYesYesNot available
Example: 100K in + 10K out$0.225$0.1125, then $0.225$0.30, then $0.60

The example row is our arithmetic from the list prices: 0.1M × $1.50 plus 0.01M × $7.50 = $0.225. Grounding with Google Search includes 5,000 free requests a month, then costs $14 per 1,000. Google also lists Batch, Flex and Priority tiers, which are not shown here.

Gemini 4 Hydrogen benchmarks: how does it score?

Scores Google DeepMind published at launch for Hydrogen and two rival models in its class. We have not run them ourselves.

SWE-sweep (agentic coding)

Gemini Hydrogen
13.5%
Claude Sonnet 5.5
3.1%
GPT-6.1 Terra
5.8%

Terminal-Bench 4.1 (cheating-adjusted)

Gemini Hydrogen
57.0%
Claude Sonnet 5.5
63.9%
GPT-6.1 Terra
not listed

Agents' Last Exam v1.1

Gemini Hydrogen
71.6%
Claude Sonnet 5.5
60.4%
GPT-6.1 Terra
69.8%

AutomationBench-AA

Gemini Hydrogen
76.1%
Claude Sonnet 5.5
71.8%
GPT-6.1 Terra
60.2%

Vending-Bench 2 (bank balance)

Gemini Hydrogen
$14,549
Claude Sonnet 5.5
$11,165
GPT-6.1 Terra
$9,107

OSWorld-2.1 (offline subset)

Gemini Hydrogen
77.6%
Claude Sonnet 5.5
83.9%
GPT-6.1 Terra
74.2%

MKA (multimodal knowledge agent)

Gemini Hydrogen
53.4%
Claude Sonnet 5.5
52.0%
GPT-6.1 Terra
43.8%

Terminal-Bench-Science 0.1

Gemini Hydrogen
51.4%
Claude Sonnet 5.5
59.9%
GPT-6.1 Terra
58.3%

HLE-Diamond (reasoning)

Gemini Hydrogen
58.8%
Claude Sonnet 5.5
53.3%
GPT-6.1 Terra
47.3%
BenchmarkGemini HydrogenClaude Sonnet 5.5GPT-6.1 Terra
SWE-sweep (agentic coding)13.5%3.1%5.8%
Terminal-Bench 4.1 (cheating-adjusted)57.0%63.9%not listed
Agents' Last Exam v1.171.6%60.4%69.8%
AutomationBench-AA76.1%71.8%60.2%
Vending-Bench 2 (bank balance)$14,549$11,165$9,107
OSWorld-2.1 (offline subset)77.6%83.9%74.2%
MKA (multimodal knowledge agent)53.4%52.0%43.8%
Terminal-Bench-Science 0.151.4%59.9%58.3%
HLE-Diamond (reasoning)58.8%53.3%47.3%

Vendor-reported numbers on Google's own setups. Google also reports 51 on the Artificial Analysis Intelligence Index at the default high level (54 at xhigh), against 50 for Gemini 4 Argon. Treat all of it as a direction, then run your own evals.

What changed from Gemini 3.8 Flash?

Google pitches the new model as a big step up in both speed and intelligence over 3.8 Flash.

Faster finishes

Google says tasks finish 2.5 times faster at the default high setting; partners report 2x to 2.8x faster runs with 12% to 23% fewer tokens.

Stronger agentic coding

Terminal-Bench 4.1 more than tripled, from 18.2% to 57%. Google adds that it batches file reads and searches in parallel.

Fewer made-up claims

Google reports fabricated claims cut by more than 5x, and cleaner recovery when a tool call fails or a search comes back empty.

New xhigh thinking level

Gemini Hydrogen adds xhigh above low, medium and high (the default), so you only pay for the reasoning a task needs.

Built to pair with Carbon

Orchestrator in Google Antigravity switches between Carbon and Hydrogen in one session, reaching near-Carbon success on Terminal-Bench 4.0 with 24% less quota.

Tighter agent safety

Against 3.8 Flash, Google reports up to 81% less overreach in agent tasks and 12x fewer misaligned actions in tool-use and coding evals. It is also more than 2.3x more resilient on the Gray Swan prompt-injection benchmark.

Where can you use Gemini Hydrogen?

Google rolled Gemini Hydrogen out globally on launch day across developer, business and consumer products.

01

Gemini API and AI Studio

Call gemini-4-hydrogen through the Gemini API from Google AI Studio or Android Studio. Unlike Carbon and Argon, it is offered on the free tier as well as the paid tier.
02

Gemini for business

Enterprises can deploy it via the API or model drop-down in Gemini for business, in Antigravity, and in Gemini in Sheets on eligible plans.
03

Gemini app, Search and Sheets

It is in the Gemini app model drop-down, and free users reach it through auto model selection when it fits the prompt. It powers Google Search for all users worldwide; Gemini in Sheets starts with Google AI Pro and Ultra.
04

Google Antigravity

Paid Google AI Pro and Ultra subscribers can select the new model directly, or use Orchestrator, which pairs it with Carbon or Argon depending on the plan.

What to check before you switch

Details from Google's pricing page and model card that affect what Gemini Hydrogen costs and how it behaves.

Do
  • Pick the thinking level per task; high is the default, xhigh costs more tokens
  • Use background long decode when you need more than 65K output tokens
  • Budget for thinking tokens, which are billed at the output rate
  • Re-check your 3.8 Flash bill: its price doubles on January 1, 2027
Don't
  • Plan on explicit context caching; only implicit caching is supported
  • Send private data on the free tier, where content is used to improve Google's products
  • Expect knowledge past March 2026 without grounding or search
  • Assume zero refusals: Google's card shows more unjustified refusals than 3.8 Flash

Can you run Gemini Hydrogen on Synexa?

No. Synexa does not serve this model or any other Gemini language model, and it offers no chat LLMs; its only text outputs are LLaVA-13B image answers, Whisper speech transcripts and an NSFW image-check label. For the language side, use the Gemini API or Gemini for business.

Synexa covers the media half of an AI product through one REST API. One key gives you image models such as Nano Banana Pro at $0.10 a request, music with Lyria 2 at $0.10 a request, plus video, speech, upscaling and image-to-3D models such as Hunyuan3D 3.1 at $0.16 a request. A common pattern: let the language model plan and write the prompt, then send it to Synexa to render the asset. Prices are on each model page in Explore, new accounts start with $0.10 of free credit, and the quickstart shows the first call.

  • Pay per request, no subscription
  • Python and Node clients, plus plain HTTP
  • Image, video, audio, upscaling and 3D in one catalog

Synexa models to pair with a Gemini Hydrogen agent

Flat per-request prices as listed on Synexa model pages, November 2026.

Product or marketing image from a written brief

Synexa model
Nano Banana Pro
Price
$0.10 / request

Image or edit with sharp, readable text

Synexa model
GPT Image 2
Price
$0.10 / request

30-second music clip from a text prompt

Synexa model
Lyria 2
Price
$0.10 / request

Textured 3D model from one image

Synexa model
Hunyuan3D 3.1
Price
$0.16 / request

Fast draft 3D mesh

Synexa model
TRELLIS
Price
$0.02 / request

Upscale or restore a photo

Synexa model
Real-ESRGAN
Price
$0.0015 / request

Transcribe a voice note

Synexa model
Whisper v3
Price
$0.004 / request
Job your agent hands offSynexa modelPrice
Product or marketing image from a written briefNano Banana Pro$0.10 / request
Image or edit with sharp, readable textGPT Image 2$0.10 / request
30-second music clip from a text promptLyria 2$0.10 / request
Textured 3D model from one imageHunyuan3D 3.1$0.16 / request
Fast draft 3D meshTRELLIS$0.02 / request
Upscale or restore a photoReal-ESRGAN$0.0015 / request
Transcribe a voice noteWhisper v3$0.004 / request

Video models such as Veo 3.1 are also in the catalog; check each video model page for its current price.

Is Gemini Hydrogen worth it over Carbon or 3.8 Flash?

For most everyday agent and coding work, yes. Gemini Hydrogen costs 37.5% of Carbon's or Argon's list price after December 1, 2026 ($1.50 / $7.50 against $4 / $20), and Google reports it slightly ahead of Argon on the Artificial Analysis index. Keep Carbon for deep architecture decisions and subtle bugs, which is how Google itself splits the work in Orchestrator.

Against 3.8 Flash the sums change over time. Until December 31, 2026, 3.8 Flash is half the price; from January 1, 2027 both cost the same, so the faster, stronger model wins on value. Good fits for Gemini Hydrogen: repository coding, data and SQL agents, long PDF and video analysis. If your app also produces images, music or 3D assets, see how a dedicated media API compares in our WaveSpeed AI alternative page, or read about another new model, Claude Fable 5.2, and check per-request prices on pricing.

Related pages

More model explainers and API comparisons.

Qwen4-Max Preview
Alibaba's next flagship Qwen model, in a two-week preview: confirmed specs, access, prices and gaps.
Claude Fable 5.2
Anthropic's most capable generally available model: specs, pricing, access and API gotchas.
Synexa vs WaveSpeed
Synexa compared with WaveSpeedAI: per-model prices, free credits, catalog and when each one fits.
Explore Synexa models
Browse the image, video, audio and 3D models you can run with one API key.
Synexa pricing
Per-request prices for the models on Synexa, with no subscription.

Frequently asked questions

What is Gemini Hydrogen?

Gemini Hydrogen, officially Gemini 4 Hydrogen, is Google's fast and efficient Gemini 4 language model, released November 12, 2026. It is built for everyday coding, agent workflows and multimodal knowledge work, reads text, images, audio, video and PDFs, and holds a 1M-token context. It sits below the frontier models Gemini 4 Carbon and Argon and is adapted from Carbon.

How much does Gemini Hydrogen cost?

On the paid Gemini API tier it costs $1.50 per million input tokens and $7.50 per million output tokens, flat across the full 1M context, as of November 2026. Thinking tokens are billed as output. Context caching costs $0.075 per million tokens. A call with 100K tokens in and 10K out costs about $0.225.

Is Gemini Hydrogen free?

Partly. The Gemini API free tier includes it at no charge, though free-tier content is used to improve Google's products. In the Gemini app, free users get it through auto model selection when it fits the prompt, and Google Search uses it for all users. Google AI Pro and Ultra subscribers can also use it in Google Antigravity and Gemini in Sheets.

Is Gemini 4 Hydrogen the same as Gemini 4 Flash?

Not by name. Before launch, reports described Hydrogen as an internal fast Gemini 4 Flash model, but Google released it as Gemini 4 Hydrogen and its launch posts use no Flash label. It fills the role Flash models played: Google pitches it as the upgrade from Gemini 3.8 Flash, which its docs now call the previous-generation Flash model.

What model ID does the API use?

The Gemini API model ID is gemini-4-hydrogen. You can try it in Google AI Studio and call it from the Gemini API, Android Studio or Gemini for business. Google's model list gives a 1M-token context window and up to 65K output tokens per normal response, or up to 1M output tokens with background long decode.

How does Hydrogen compare with Gemini 4 Carbon?

Carbon is Google's most capable reasoning model and Hydrogen is the fast executor. Hydrogen costs $1.50 / $7.50 per million tokens against Carbon's $4 / $20 from December 1, 2026, and Carbon is not on the free tier. Google suggests pairing them: Carbon plans and repairs, Hydrogen writes edits and runs tests, which Orchestrator in Antigravity does automatically.

What is the knowledge cutoff?

Google's model card gives a knowledge cutoff of March 2026, and notes that some domains may be less up to date. For anything newer, turn on grounding with Google Search, which includes 5,000 free search requests a month on the paid tier and then costs $14 per 1,000 requests, subject to a daily cap of 1,500.

Can I use Gemini Hydrogen through Synexa?

No. Synexa does not serve any Gemini model or any other chat language model. Synexa is a pay-per-request API for generative media: images, video, music and speech, upscaling and image-to-3D. Many teams call a language model for planning and writing, then call Synexa to render the image, clip, track or 3D model. New Synexa accounts get $0.10 of free credit.

Pair Gemini Hydrogen with a media API

Use Gemini Hydrogen for the reasoning, then generate the images, video, music and 3D models it plans on Synexa. Pay per request, start with $0.10 free credit.

Explore Synexa models