G
GOOGLE / gemini-4-hydrogen

Gemini 4 Hydrogen

Google’s stable multimodal model, available since November 12 for everyday coding, debugging and tool-using agents. It accepts text, images, audio, video and PDFs and generates text; it does not generate images or speech.

01 / SUMMARY

The essential

ORIENTATIONCoding and agent execution

Primary use stated or inferred cautiously from official documentation.

CONTEXT1,048,576 tokens shared by input and output

Maximum input capacity when published by the source.

EXIT65,536 synchronous tokens; up to 1,048,576 with long decode, within the shared context

Documented maximum generation limit.

ACCESSGemini API · Google AI Studio

Verified availability channels.

MODEL TYPEText and reasoning · Code · Vision and documents

Primary functional family and verified specialties.

LICENSEProprietary

Declared terms for API access, model weights, or self-hosting.

02 / TECHNICAL SHEET

Limits and integration

API IDgemini-4-hydrogen
Model typeText and reasoning · Code · Vision and documents
Access modelAPI with free and paid tiers
LicenseProprietary
DeploymentHosted API
Release12 NOV 2026
Knowledge cutoffNot published
EntranceText · Image · Video · Audio · PDF
ExitText
Context window1,048,576 tokens shared by input and output
maximum output65,536 synchronous tokens; up to 1,048,576 with long decode, within the shared context
Reasoninglow · medium · high (default) · xhigh
Published toolsFeatures · Code execution · Computer use · Web search · File search
Structured OutingsSí
Batch processingYes · 50% discount
Prompt cacheImplicit only; explicit caching is unsupported
Fine-tuningNot published
Verified platformsGemini API · Google AI Studio · Google Antigravity
BEST SUITED FOR

Resolve repository issues, run tasks with tools and synthesize visual documents. Four reasoning levels let you adjust time and consumption; high is the default. It can work alone or alongside more capable models.

WORTH MONITORING

The million-token window is unified: input and output share its capacity. Default synchronous output is 65,536 tokens; long decode requires its specific workflow. Live API, audio or image generation and explicit caching are unsupported. Validate tools and external data before executing actions.

03 / CAPABILITIES

What it can do

01

Orientation

Resolve repository issues, run tasks with tools and synthesize visual documents. Four reasoning levels let you adjust time and consumption; high is the default. It can work alone or alongside more capable models.

02

Context

1,048,576 tokens shared between input and output · 65,536 synchronous tokens; up to 1,048,576 with long decode, within the shared context

03

Tools and integration

Function calling · Code execution · Computer use · Web search · File search

04

Access

Gemini API · Google AI Studio

Input modalities
TextImageVideoAudioPDF
04 / PRICES

Documented cost

ConceptWorthUnit/condition
Standard input$1.50 / 1M tokensStandard API rate
Cached input$0.075 / 1M tokensReading reused prefixes
Cache write or storageNot publishedThe condition varies by provider
Standard output$7.50 / 1M tokensMay include reasoning tokens
Batch input$0.75 / 1M tokensAsynchronous processing
Batch output$3.75 / 1M tokensAsynchronous processing

Standard paid Gemini API pricing is USD 1.50 for input and USD 7.50 for output per million tokens, including reasoning. Batch and Flex cost USD 0.75/3.75; Priority has different rates. Implicit cached input costs USD 0.075. Tools may add charges. The free tier has its own limits and data-use terms.

i Prices change and may depend on level, region or context length. Check the source before making a decision.

05 / EVALUATIONS

How to read the results

“

Google’s model card reports cheating-adjusted Terminal-Bench 4.1, HLE-Diamond, MKA and Agents’ Last Exam v1.1. Versions, metrics and configurations matter: these figures are not Terminal-Bench 4.0 results or Inferama’s own tests. Results for Carbon, Argon or Orchestrator are not attributed to Hydrogen.

BenchmarkResultMetricSource
Terminal-Bench 4.157,0 %Cheating-adjusted successView source ↗
HLE-Diamond58,8 %ScoreView source ↗
MKA53,4 %Climb WSR accuracyView source ↗
Agents’ Last Exam v1.171,6 %ScoreView source ↗

Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.

06 / VERSIONS

Chronology

gemini-4-hydrogen

General availability in Gemini API; Batch is a consumption mode of the same model.

ANALYSIS

Related analysis

07 / SOURCES

Traceability