Google launches Gemini 4 Hydrogen, a faster model for AI agents and coding
Image generated by the Inferama Editorial Team
01

Gemini 4 Hydrogen joins the Gemini 4 family

Google announced Gemini 4 Hydrogen on November 12, 2026, adding a third model to the Gemini 4 family alongside Gemini 4 Argon and Gemini 4 Carbon. Google positions Argon and Carbon for complex reasoning and other demanding work, while Hydrogen is designed to complete tasks quickly in AI-agent workflows and everyday coding.

The distinction is about the work each model is meant to handle. Hydrogen targets speed and efficiency, rather than the deepest or longest-running reasoning. Google began rolling it out through its consumer products and developer tools on its announcement date, in contrast with the more restricted initial access for Argon and Carbon. Availability can still depend on the product, plan and region; access in one channel does not establish access everywhere.

02

What Google’s speed claim measures

Google says Hydrogen completes tasks 2.5 times faster than Gemini 3.8 Flash at Hydrogen’s default reasoning level. The company also reports a Terminal-Bench 4.1 score of 57% for Hydrogen, compared with 18.2% for Gemini 3.8 Flash. These figures describe a particular comparison and benchmark, not a universal speed advantage across every coding project or agent setup.

Google presents speed as part of a broader efficiency goal, including lower token use and recovery from tool or search failures. Those claims matter for workflows that involve multiple steps, but they do not by themselves show that every task will finish faster or that generated code will be correct. The 2.5× figure is a manufacturer-reported result, and independent replication is not established here.

Google’s Terminal-Bench 4.1 comparison

Google’s reported results for Gemini 4 Hydrogen and Gemini 3.8 Flash. The speed figure is the company’s comparison at the default reasoning level.

ModelTerminal-Bench 4.1 scoreReported task speed
Gemini 3.8 Flash18.2%Baseline
Gemini 4 Hydrogen57%2.5× faster than Gemini 3.8 Flash
03

Four reasoning settings for different workloads

Hydrogen offers four reasoning settings: low, medium, high and xhigh. High is the default, and xhigh is a new option in this model. Users can choose a setting to adjust how much reasoning the model applies, with the aim of balancing response time and cost against the demands of a task.

Google CEO Sundar Pichai described a trade-off between reasoning depth, waiting time and cost. In practice, a setting is a control over the model’s inference behavior, not a guarantee of a particular answer quality, response time or bill. A lower setting does not ensure valid code or accurate output, and a higher setting does not ensure that the model will solve a difficult task. Teams should compare settings on representative workloads and check results before acting on them.

04

Benchmarks and comparisons with other models

Google reports that Hydrogen scores 51 points on the Artificial Analysis Intelligence Index at its standard reasoning level, compared with 50 points for Gemini 4 Argon. Google also says Hydrogen leads in seven of ten comparisons with Claude Sonnet 5.5 and GPT-6.1 Terra, including HLE-Diamond and MKA. These are Google-reported comparisons, and the selected tests do not establish that Hydrogen is better for every use case.

The same comparison has exceptions. Google’s published figures show Claude Sonnet 5.5 ahead on Terminal-Bench 4.1 and OSWorld-2.1, which assess terminal-based coding and computer use. Results can vary with the benchmark version, model settings, tools and evaluation conditions. Until independent evaluations using comparable configurations establish otherwise, treat these figures as evidence about the reported tests rather than a general ranking of the models.

Selected benchmark results reported by Google

Google’s comparison of Gemini 4 Hydrogen, Claude Sonnet 5.5 and GPT-6.1 Terra. N/D means Google did not provide a figure in this comparison.

BenchmarkGemini 4 HydrogenClaude Sonnet 5.5GPT-6.1 Terra
HLE-Diamond58.8%53.3%47.3%
SWE-sweep13.5%3.1%5.8%
Agents’ Last Exam71.6%N/D69.8%
05

Availability, API pricing and access channels

Google lists API prices of $1.50 per million input tokens and $7.50 per million output tokens for Hydrogen. The reported rates apply across context lengths up to one million tokens. These are API prices; the way people access Hydrogen in consumer products may depend on plan features, automatic model selection, usage limits and regional rollout.

Google is rolling Hydrogen out through Google AI Studio, the Gemini API, Android Studio and Firebase AI Logic. It is also being integrated into Google Search’s AI Mode. In the Gemini app, Google’s plans page lists Hydrogen with Google AI Plus, while free users may receive it through automatic model selection. Google’s plans page lists expanded access to models and developer tools for paid tiers, but those descriptions do not mean every feature or usage limit is identical across countries. Check the product and plan available to your account before relying on access for a workflow.

Search integration is intended to help with complex or poorly phrased queries and to produce more useful results. That product feature is separate from API access: availability in Search does not imply that a user can call the model through the API, and API availability does not establish identical access in every Gemini feature. The model’s rollout and the precise limits may differ by market and product.

Price and access at a glance

Google’s announced API pricing and the reported access channels. Consumer-product availability and usage limits may vary by plan and region.

ChannelPrice or accessNotes
Gemini API$1.50 per million input tokens; $7.50 per million output tokensGoogle lists access across API tiers
Gemini appIncluded through eligible plans; free users may receive it through automatic selectionGoogle AI Plus lists Hydrogen; features and limits can vary
Google SearchRolling out through AI ModeAvailability depends on product rollout and market
06

Keep exploring

Sources consulted

03

Corrections and transparency

If you spot incorrect or outdated information, send us a correction with the page and source we should review.

Submit a correction