Google

Gemini 4 Hydrogen

Google's fast Gemini 4 model for everyday agent, coding and multimodal work. It replaces Gemini 3.8 Flash.

adapted from gemini-4-carbontext inputtext output

Replaces Gemini 3.8 Flash: input $0.7875 → $1.575 per 1M (+100%), output $3.9375 → $7.875 per 1M (+100%), same context window, max output 66K → 1M.

Run this model

google/gemini-4-hydrogen

Use an endpoint supported by this model.

Get a key for this setup →
Thinking guide →
const response = await fetch('https://api.minirouter.sh/v1/chat/completions', {  method: 'POST',  headers: {    Authorization: `Bearer ${process.env.MINIROUTER_KEY}`,    'Content-Type': 'application/json',  },  body: JSON.stringify({    model: 'google/gemini-4-hydrogen',    messages: [{ role: 'user', content: 'Why is the sky blue?' }],    max_tokens: 1024,  }),}) const data = await response.json()console.log(data.choices[0].message.content)
Read docs →

max_tokens covers thinking plus the answer, so leave room for both. High effort is the default; xhigh waits about 40 seconds before the answer starts.

Model overview

Input / 1M
$1.575
All-in MiniRouter rate
Output / 1M
$7.875
All-in MiniRouter rate
Context
1M
1,000,000 tokens
Max output
1M
tokens
Effort
low → xhigh
high by default, always thinking
Intelligence Index
54
Artificial Analysis, xhigh effort
Announced
12 Nov 2026
publisher release

Compare text-model alternatives or estimate token costs for your workload.

Publisher-reported, plotted by MiniRouter

Ahead on agents and workflows, behind on terminal work.

Hydrogen leads six of these nine rows from Google's model card. Claude Sonnet 5.5 leads Terminal-Bench 4.1, OSWorld-2.1 and Terminal-Bench-Science. MiniRouter plotted the table and did not run the evaluations.

Long-running agents

Finishes more of a five-hour agent exam.

122 long agent tasks, each with a five-hour window. Google ran Hydrogen in its Antigravity CLI harness; the other two scores are their publishers' own.

Gemini 4 Hydrogen
71.6%Highest thinking
GPT-6.1 Terra
69.8%Self-reported
Claude Sonnet 5.5
60.4%Self-reported

Agents' Last Exam v1.1

0 to 100%

  • Gemini 4 Hydrogen71.6%
  • Claude Sonnet 5.560.4%
  • GPT-6.1 Terra69.8%
Source: Google model card ↗ · Evaluation methods ↗. Results as of November 2026.
View chart data
Agents' Last Exam v1.1 scores
ModelScore
Gemini 4 Hydrogen71.6%
Claude Sonnet 5.560.4%
GPT-6.1 Terra69.8%

Eight more evaluations

Workflows, coding, computer use and research

  • Gemini 4 Hydrogen
  • Comparison models

Business workflows

AutomationBench-AA

  • Gemini 4 Hydrogen76.1%
  • Claude Sonnet 5.571.8%
  • GPT-6.1 Terra60.2%

0 to 100%

Expert reasoning

HLE-Diamond

  • Gemini 4 Hydrogen58.8%
  • Claude Sonnet 5.553.3%
  • GPT-6.1 Terra47.3%

0 to 100%

Agentic coding

SWE-sweep

  • Gemini 4 Hydrogen13.5%
  • Claude Sonnet 5.53.1%
  • GPT-6.1 Terra5.8%

0 to 100%

Terminal coding

Terminal-Bench 4.1

  • Gemini 4 Hydrogen57.0%
  • Claude Sonnet 5.563.9%
  • GPT-6.1 Terra—Not reported

0 to 100%

Computer use

OSWorld-2.1 · offline subset

  • Gemini 4 Hydrogen77.6%
  • Claude Sonnet 5.583.9%
  • GPT-6.1 Terra74.2%

0 to 100%

Scientific research

Terminal-Bench-Science 0.1

  • Gemini 4 Hydrogen51.4%
  • Claude Sonnet 5.559.9%
  • GPT-6.1 Terra58.3%

0 to 100%

Knowledge work

MKA

  • Gemini 4 Hydrogen53.4%
  • Claude Sonnet 5.552.0%
  • GPT-6.1 Terra43.8%

0 to 100%

Chart reading

Chartography

  • Gemini 4 Hydrogen66.6%
  • Claude Sonnet 5.561.6%
  • GPT-6.1 Terra63.2%

0 to 100%

Hydrogen at its highest thinking setting, except Terminal-Bench 4.1 at high. Other models' scores are their providers' own or from public leaderboards, at maximum reasoning where available. Source: Google model card ↗ · Evaluation methods ↗. Results as of November 2026.
View chart data
Values from Google's Gemini 4 Hydrogen model card
BenchmarkGemini 4 HydrogenClaude Sonnet 5.5GPT-6.1 Terra
Agents' Last Exam v1.171.6%60.4%69.8%
AutomationBench-AA76.1%71.8%60.2%
HLE-Diamond58.8%53.3%47.3%
SWE-sweep13.5%3.1%5.8%
Terminal-Bench 4.157.0%63.9%Not reported
OSWorld-2.1 · offline subset77.6%83.9%74.2%
Terminal-Bench-Science 0.151.4%59.9%58.3%
MKA53.4%52.0%43.8%
Chartography66.6%61.6%63.2%
Choose an effort

Xhigh adds three points and 26 seconds.

High is Google's default and starts answering in about 15 seconds. Xhigh is new with Hydrogen: it scores three points higher and waits about 41 seconds. Low answers in under six seconds at roughly half the score.

Intelligence Index v5

Low
27
Medium
41
High
51
Xhigh
54

Wait before the answer starts

Seconds to first answer token

0 to 60 seconds

  • Low5.8
  • Medium12.3
  • High15
  • Xhigh41.1
Artificial Analysis ↗, measured on Google's API and read November 12, 2026. These are not MiniRouter route measurements.
View chart data
Gemini 4 Hydrogen by effort, measured by Artificial Analysis
EffortIntelligence IndexFirst answer tokenOutput speed
Low275.8 s186 tokens/s
Medium4112.3 s177 tokens/s
High5115.0 s236 tokens/s
Xhigh5441.1 s198 tokens/s
Publisher pricing

One rate across the whole million-token window.

Google lists Hydrogen 25% below Carbon and Argon today, with no step up for long prompts and no dated increase. Carbon and Argon double on December 1. The price panel above shows what you pay on MiniRouter.

Google list, per 1M tokens

Input
$1.50Carbon and Argon: $2.00
Output
$7.50Carbon and Argon: $10.00
Cached input
$0.075Carbon and Argon: $0.10
List prices from Google's pricing page ↗, read November 12, 2026, before MiniRouter's fee. Output includes thinking tokens.
View chart data
Google list prices per million tokens for Gemini 4 Hydrogen, Carbon and Argon
RateHydrogenCarbon and Argon through November 30, 2026Carbon and Argon from December 1, 2026
Input$1.50$2.00$4.00
Output$7.50$10.00$20.00
Cached input$0.075$0.10$0.20
Before you switch

Four behaviours to plan for.

Google describes these in its Hydrogen guide. Carbon and Argon requests need no other change than the model ID.
  1. [ 01 ]

    The output cap counts thinking

    max_tokens covers thinking plus the visible answer, so a small cap can be spent before the answer starts.

  2. [ 02 ]

    Thinking cannot be turned off

    Choose low, medium, high or xhigh. Google's default is high, up from medium on Gemini 3.8 Flash.

  3. [ 03 ]

    Sampling parameters are ignored

    Google ignores temperature, top_p and top_k on Hydrogen.

  4. [ 04 ]

    One reply holds up to 65,536 tokens

    That is Google's limit for thinking plus answer in a single response.

Route

One route, standard clients.

Authenticate with a MiniRouter bearer key. Caller-provided upstream credentials are not used on this route.
Base URLhttps://api.minirouter.sh/v1
Model IDgoogle/gemini-4-hydrogen
Endpoints/v1/chat/completions/v1/messages/v1/responses
Parametersmax_tokens, temperature, stop, tools, tool_choice, reasoning, include_reasoning
Structured outputNot enforced; a requested format is dropped and the answer is plain text.
ModalitiesInput: text · Output: text

Configured providers. Priority follows the live catalog order.

ProviderPriorityInput / 1MOutput / 1MParametersAccess
vercel
1$1.575$7.875max_tokens, temperature, stop, tools, tool_choice, reasoning, include_reasoningMiniRouter balance

Every model, one key.