Google introduced Gemini 4 Hydrogen on November 12, 2026, and put it everywhere at once. It is the fast model in the Gemini 4 family, built for everyday coding and agentic work, and it reached the Gemini app, Google Search, AI Studio, the Gemini API, Gemini for business and Google Antigravity on the same day. This article sets out what Google announced, how Hydrogen compares with Carbon and Argon, what the benchmarks say, what it costs, and where you can use it. Every date, price and score below comes from Google, GitHub or the named benchmark owners.
What Google announced on November 12
Google announced Gemini 4 Hydrogen on November 12, 2026, as a fast model for everyday coding and agentic work. It is the third Gemini 4 model, after Carbon and Argon, and it reaches every Google surface at once: AI Studio, the Gemini API, the Gemini app, Google Search and Antigravity.
The announcement came from Koray Kavukcuoglu, senior vice president at Google DeepMind and chief AI architect at Google, in a post on the Google blog titled ‘Introducing Gemini 4 Hydrogen for Agentic Workflows.’ The post frames Hydrogen as the execution model of the family. Carbon and Argon handle deep, long-horizon reasoning on hard problems. Hydrogen handles rapid execution across everyday coding and agent workflows. Google also used the launch to bring Gemini 4 to everyone, not only to paid subscribers. The model rolled out globally on the same day across developer, enterprise and consumer surfaces. That matters because the two earlier Gemini 4 models stayed behind paywalls. Argon went to paid AI Pro and Ultra subscribers, and Carbon went to Ultra subscribers in AI Mode. Hydrogen is the first Gemini 4 model that free users meet in Search and in the Gemini app. Our Google Gemini 3 guide covers the generation Hydrogen replaces, and our November 2026 core update report covers the same week in Search.
How Hydrogen compares with Carbon and Argon
Hydrogen is the fast, low cost member of the Gemini 4 family. Carbon is the strongest model for deep reasoning, and Argon sits between the two. Hydrogen beats Argon on the Artificial Analysis Intelligence Index at its default setting, 51 to 50, and reaches 54 at its highest thinking level.
Google positions the three models by job, not only by size. Carbon is the frontier model for architectural judgment, hard bugs and long research. Argon is a paid model for Pro and Ultra subscribers. Hydrogen is the model for work that repeats all day: writing code, running tests, checking changes and running agents. The Artificial Analysis Intelligence Index scores Hydrogen at 51 against Argon at 50 at the default high thinking level, and at 54 at the xhigh level. The more useful comparison is speed against the model Hydrogen replaces. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its high default setting. It more than triples the Terminal-Bench 4.1 score, from 18.2 percent to 57 percent, and it cuts fabricated claims by more than five times. It also adapts when a tool call fails or a search returns empty, instead of stopping or guessing. The model card says Hydrogen is adapted from Carbon. Readers who follow rival models can compare this with our GPT-5.1 report and our Claude Sonnet 4.5 report.
Four thinking levels
Hydrogen offers four thinking levels: low, medium, high and a new xhigh tier. High is the default. Each level trades a little speed for more reasoning depth, so a simple edit costs less time and fewer tokens than a hard debugging task. The steps scale in a smooth, predictable way.
The thinking levels are the control that makes Hydrogen fit different jobs. Low suits quick edits and short answers. Medium suits routine coding. High, the default, suits most agent work. Xhigh is new, and it is the level that lifts the index score to 54. Google says the levels scale in a smooth, predictable progression, so users pay and wait only for the reasoning depth a task needs. Across the levels, Hydrogen matches or beats rival models in its class while often finishing faster and using fewer output tokens, according to Google. Developers pick the level in AI Studio and in the API. In the Gemini app and in Search, Google picks the level behind the scenes.
Coding and agentic workflows
Hydrogen is built for repository coding, debugging and multi-step agent work. Google says it finishes tasks 2.5 times faster than Gemini 3.8 Flash and more than triples its Terminal-Bench 4.1 score, from 18.2 percent to 57 percent. It also cuts fabricated claims by more than five times.
The coding claims rest on tests Google chose, so they need that label. On SWE-sweep, where the model has to find and repair bugs without breaking existing code, Google says Hydrogen beats similar-tier models with focused, surgical edits. It batches file-read and search calls in parallel instead of inspecting files one by one, which matters in large repositories. In early testing across complex graphs, SQL and data-heavy enterprise work, Google’s partners saw Hydrogen run 2 times to 2.8 times faster than 3.8 Flash while cutting token use by 12 percent to 23 percent. On Agent Arena, a live benchmark, Hydrogen ranks fifth on multi-step tasks such as deep research, data analysis and broad codebase navigation. For deep domain reasoning it leads its class on HLE-Diamond at 58.8 percent, ranks first overall on Vals Tax Agent Bench at 52.3 percent, ties for first on Vals Legal Research Bench at 55.8 percent, and leads its class on AutomationBench-AA. These are vendor-selected benchmarks, not an independent verdict.
Multimodal knowledge work
Hydrogen pairs 1 million input tokens with up to 1 million output tokens, so it can read long PDFs, spreadsheets, charts, audio and recorded video, then return a cited report. Google says it leads its class on the Multimodal Knowledge Agent benchmark with a score of 53.4 percent.
The long context window is the other half of the Hydrogen story. A 1 million-token input window lets the model take in dense files in one pass: a long PDF report, a spreadsheet, a chart deck, an audio recording or a recorded video. It can then return a polished, cited report or a structured deliverable. Google says Hydrogen beats other models on the Multimodal Knowledge Agent benchmark at 53.4 percent for end-to-end agentic knowledge work over complex visual files, and scores 71.6 percent on Agents’ Last Exam v1.1 for end-to-end professional workflows. The practical use is clear. A researcher can hand the model a pile of source material and get back a report with citations. A team can feed it meeting audio and spreadsheets and get back a structured summary. The output limit matters too: up to 1 million output tokens means long reports do not get cut short.
What changed in Google Search
Google put Hydrogen into Search for all users on November 12, 2026. Robby Stein called it the biggest model upgrade to Search yet. It reads multi-part questions better, returns more visual answers and links to a wider range of sites through a technique Google calls query fan-out.
Search is where most people will meet Hydrogen first. Google says Hydrogen understands a question even when it has several parts, typos or a very specific request. It returns answers that are more visual and easier to scan, with images, structured responses, comparison tables and custom tools. It also upgrades query fan-out, the technique that runs simultaneous searches across subtopics. That connects users to a broader range of helpful links than before, including sites they might not have found otherwise. The change applies globally to all Search users. Google’s Search post names no benchmarks for Search itself, and it says nothing about ads or clicks. The November 2026 core update started two days earlier, on November 10, so ranking and answer changes landed in the same week. The core update changes ranking; Hydrogen changes how Search builds answers.
What changed in the Gemini app
In the Gemini app, Hydrogen is the default fast model in the model menu. Google says it follows instructions more closely, calls the right tools more often and finishes everyday work with less back and forth. It also pulls context from Gmail, Google Drive and connected apps such as Airtable and Dropbox.
The Gemini app changes are about everyday reliability. Google says Hydrogen stays on track and follows instructions more closely across multi-turn conversations. Email drafts, study guides, rewrites and documents come back ready to use on the first try, because the model sticks to the tone, format and rules the user set. When comparing products or dense material, it builds clear tables and visual breakdowns without dropping key details. It also breaks a goal into steps, pulls context from uploads, Gmail, Google Drive and connected apps such as Airtable and Dropbox, and calls the right tools and skills for the job. For prototyping, Hydrogen generates clean, working layouts in fewer prompts. Google’s example is a project that turns real-time global lightning strike data into an interactive 3D globe. Users pick Hydrogen from the model menu in the Gemini app. In Google Sheets, Hydrogen starts with Google AI Pro and Ultra subscribers.
Safety and reliability
Hydrogen uses the latest Gemini 4 safety stack, with built in training and real time safeguards. Google says it cuts overreach in agentic tasks by up to 81 percent and under-service by up to 79 percent. It is also more than 2.3 times more resilient to indirect prompt injection attacks.
Safety matters more for a fast agent model than for a chat model, because agents take actions. Google says Hydrogen cuts misaligned actions and task shortcutting across tool-use and coding tests by 12 times. When a task is impossible or blocked, it is up to 2 times less likely to bypass permission boundaries, and it scores a 94 percent honesty rate on Google’s internal agent honesty test. Google also says Hydrogen keeps 99.9 percent chain-of-thought legibility with zero deceptive reasoning edits across more than 200,000 audited turns. Against indirect prompt injection, where an attacker plants a prompt in tool output, Hydrogen is more than 2.3 times more resilient on the Gray Swan IPI benchmark. For everyday use, Google tuned the safeguards to tell real biosecurity risks apart from legitimate medical or scientific questions, keeping false refusals on HealthBench Pro under 0.4 percent. The full detail sits in the Gemini 4 Hydrogen model card from Google DeepMind.
Pricing, GitHub Copilot and where to use it
Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, flat across the full 1 million token window. It is available in AI Studio, the Gemini API, the Gemini app, Google Search, Gemini for business and Antigravity, and since November 13 in GitHub Copilot.
The flat price is the part developers will notice first. The same $1.50 and $7.50 rates apply across the whole 1 million-token window, with no higher price for long context. Argon and Carbon cost $2 and $10 per million tokens until November 30, 2026, and move to $4 and $20 after that, according to pricing reports. Hydrogen has a free tier in AI Studio and in the Gemini API, with rate limits. On November 13, GitHub said Gemini 4 Hydrogen is generally available in GitHub Copilot for Pro, Pro+, Max, Business and Enterprise users, with a gradual rollout. GitHub bills it at provider list pricing under usage-based billing. GitHub also said it will retire Gemini 3.8 Flash from Copilot on December 14, 2026, with Hydrogen as the suggested replacement. In Google Antigravity, paid Pro and Ultra subscribers get Hydrogen plus Orchestrator, which switches between Carbon and Hydrogen in one session. Google says Orchestrator reached near-Carbon task success on Terminal-Bench 4.0 while using 24 percent less quota and finishing 19 percent faster. In multi-agent work, Carbon can coordinate while Hydrogen powers worker and verifier agents in separate workspaces.
Frequently asked questions
Gemini 4 Hydrogen is Google’s fast Gemini 4 model for everyday coding and agentic work, announced on November 12, 2026. It is available in the Gemini app, Google Search, AI Studio, the Gemini API and GitHub Copilot. It costs $1.50 per million input tokens and $7.50 per million output tokens.
What is Gemini 4 Hydrogen?
Gemini 4 Hydrogen is Google’s fast, low cost Gemini 4 model, announced on November 12, 2026. It is built for everyday coding and agentic workflows, and it is the first Gemini 4 model available to all users in Search and in the Gemini app.
How is Hydrogen different from Carbon and Argon?
Carbon is the strongest Gemini 4 model for deep, long-horizon reasoning. Argon sits in the middle and is for paid subscribers. Hydrogen is the fast execution model. It beats Argon on the Artificial Analysis Intelligence Index at its default setting, 51 to 50, and reaches 54 at xhigh.
How much does Gemini 4 Hydrogen cost?
Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, flat across the full 1 million-token context window. A free tier with rate limits is available in AI Studio and in the Gemini API.
Where can I use Gemini 4 Hydrogen?
Hydrogen is available in all tiers of Google AI Studio, in the Gemini API, in the Gemini app, in Google Search for all users, in Gemini for business, and for paid Google AI Pro and Ultra subscribers in Google Antigravity. Since November 13, 2026, it is also generally available in GitHub Copilot.
What are the four thinking levels?
Hydrogen offers low, medium, high and xhigh thinking levels. High is the default. Each level trades speed for reasoning depth. Xhigh is new and lifts the Artificial Analysis Intelligence Index score to 54.
Is Gemini 4 Hydrogen faster than Gemini 3.8 Flash?
Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its high default setting. It also more than triples the Terminal-Bench 4.1 score, from 18.2 percent to 57 percent, and cuts fabricated claims by more than five times.
Does Hydrogen replace Gemini 3.8 Flash?
Google presents Hydrogen as the upgrade from 3.8 Flash. GitHub will retire Gemini 3.8 Flash from Copilot on December 14, 2026, with Hydrogen as the suggested replacement. Google keeps 3.8 Flash on its own price list until December 31, 2026.
