Google DeepMind announced Gemini 4 Hydrogen on November 12, 2026, describing the model as an effort to bring "cutting-edge intelligence to everyone," with speed and efficiency tuned for everyday coding and agentic workflows. The announcement was authored by Koray Kavukcuoglu, Senior Vice President at Google DeepMind and Chief AI Architect at Google.

DeepMind positioned Hydrogen as the rapid-execution counterpart to its frontier models, Gemini 4 Carbon and Gemini 4 Argon, which the company says excel at deep, long-horizon reasoning on the hardest problems. With Hydrogen, the Gemini 4 family now reaches every tier of Google AI Studio, the Gemini API, Gemini for business, the Gemini app, and Google Search, along with paid Google AI Pro and Ultra subscribers in Google Antigravity.

On the Artificial Analysis Intelligence Index, Hydrogen surpasses Argon at its default high thinking level, scoring 51 versus 50, and reaches 54 at its highest "xhigh" thinking setting, while often completing tasks faster than models in its class. The model is priced flat across its entire 1 million-token context window, at $1.50 per million input tokens and $7.50 per million output tokens. The index is published by independent AI benchmarking firm Artificial Analysis, giving the Hydrogen-versus-Argon comparison a third-party reference point, and because pricing is flat, the same per-token rates apply no matter how much of the 1 million-token window a request uses.

Unlocking efficiencies in multi-model workloads

In real-world software development, DeepMind said, teams rely on frontier models such as Carbon and Argon for deep architectural judgment when designing systems or untangling subtle bugs, but they still need a fast, disciplined execution model for writing code, running tests, and verifying changes. Hydrogen is built to work on its own for everyday tasks, or to pair with Carbon and Argon as a high-efficiency execution engine in multi-model workflows.

To automate that collaboration, Google introduced Orchestrator in Google Antigravity. Rather than relying on a single model from start to finish, Orchestrator dynamically switches between models such as Carbon and Hydrogen within the same session. On Terminal-Bench 4.0, Orchestrator achieved near-Carbon task success rates while consuming, on average, 24% less quota and completing tasks 19% faster. Under per-token pricing, token quota is the direct spending lever in these multi-model setups, which is what makes a 24% reduction meaningful.

For workflows that require parallel agents, Carbon can act as the coordinator while Hydrogen powers worker and verifier agents that build and test changes simultaneously across isolated workspaces, helping teams ship complex projects faster without driving up token costs.

Scaling developer and enterprise workflows

Beyond multi-model setups, Hydrogen is a major standalone upgrade for teams migrating from Gemini 3.8 Flash. DeepMind said it finishes tasks 2.5 times faster at its high default setting, more than triples Terminal-Bench 4.1 scores (57% versus 18.2%), cuts fabricated claims by more than 5x, and adapts cleanly when a tool call fails or a search returns empty.

The company said Hydrogen matches or outperforms competing models in its class across core developer and enterprise workflows:

Iterative repository coding and debugging. Hydrogen handles issue resolution, debugging, rapid iteration, and codebase Q&A with focused, surgical edits, delivering cleaner, production-ready code without rewriting unrelated files. It outperforms similar-tier models on SWE-sweep, where it must discover and repair bugs without breaking existing code. To explore large repositories efficiently, the model natively batches file-read and search calls in parallel rather than inspecting files one by one. In early testing across complex graphs, SQL, and data-heavy enterprise workflows, partners saw Hydrogen run 2x to 2.8x faster while reducing token usage by 12% to 23% compared with 3.8 Flash.

Open-ended agentic execution. On multi-step tasks such as deep research, data analysis, and broad codebase navigation, Hydrogen ranks #5 on Agent Arena, arena.ai's live benchmark. For deep domain reasoning and analysis, it leads its class on HLE-Diamond (58.8%), ranks #1 overall on Vals Tax Agent Bench (52.3%), ties for #1 on the Vals Legal Research Bench (55.8%), and leads its class on AutomationBench-AA.

Multimodal knowledge work. Hydrogen pairs native 1 million-token input with up to 1 million output tokens, allowing it to synthesize dense PDFs, spreadsheets, charts, audio, and recorded video into polished, cited reports and structured deliverables. It outperforms other models on the new Multimodal Knowledge Agent (MKA) benchmark (53.4%) for end-to-end agentic knowledge work over complex visual files, and on Agents' Last Exam v1.1 (71.6%) for end-to-end professional workflows.

Web development. On tasks where speed matters to developers, such as live UI prototyping, design-to-code iteration, and interactive app building, Hydrogen generates more functional, feature-complete layouts in fewer prompts.

DeepMind highlighted several demonstrations built with the model. Geminisland uses Hydrogen for on-the-fly 3D coding, letting anyone add to a shared voxel island from a single prompt, generating custom structures live on the map that players can walk around and explore in first person. Paired with a new Gemini front end design skill in Google Antigravity, Hydrogen built an e-commerce storefront in one shot, complete with a WebGL2 topographic shader, pinned four-stage scrollytelling, a thermal split-slider, a 3D wireframe spec card, and a live finish configurator.

In Google Antigravity, Hydrogen turned Google Sheets into an animated pixel canvas, shrinking thousands of cells by combining Python, native =LET(...) formulas, and conditional formatting to paint and animate full-motion scenes directly inside a live spreadsheet. Starting from a YouTube keynote video and a design template, the model generated a multi-slide presentation directly in Google Slides, then ran an autonomous audit-and-fix loop to catch overlapping text and clean up every slide layout.

With Gemini in Google Sheets, Hydrogen can help teams rapidly turn financial data into interactive models that adjust dynamically based on various forecasting scenarios, generating outputs more than 30% faster than previous models in Sheets canvas.

Adaptable speed and intelligence across four thinking levels

Hydrogen offers four thinking levels — low, medium, high (the default), and a new xhigh tier — spanning a wide range of options across quality, latency, and cost. DeepMind said the levels scale in a smooth, predictable progression so users only wait and pay for the reasoning depth a task requires. Across thinking levels, the model matches or outperforms some competing models in its class while often finishing tasks faster and using fewer output tokens.

Everyday assistance in the Gemini app and Search

For consumers, Gemini 4 Hydrogen is designed to make it easier to get things done, staying on track and following instructions more closely across multi-turn conversations. In the Gemini app, DeepMind said the model is better at following user guidance, calling the right tools, and producing a reliable output with less back-and-forth.

In writing, research, and everyday work, Hydrogen sticks closely to the user's tone, format, and specific rules, so email drafts, study guides, rewrites, and documents come back ready to use on the first try. When comparing products or synthesizing dense material, it pulls together clear tables and visual breakdowns without dropping key details.

The model breaks goals down step by step, pulling relevant context from user uploads, Google apps such as Gmail and Google Drive, and Connected Apps like Airtable and Dropbox. Google said that when a task starts in Gemini, the model more frequently calls the right tools and skills for the job.

For prototyping and iteration, Hydrogen generates clean, working layouts in fewer prompts, making it easy to test and refine interactive projects quickly. Examples cited by Google include an interactive 3D globe and Web Audio music engine driven by real-time global lightning strike data, mapping every bolt around the world into visual ripples and live ambient music; a WebGL2 water and glass visualizer built from pure math and SDF textures, simulating real-time light refraction, reflection, dispersion, and absorption; and an archival sizzle reel about the element hydrogen, created with a custom video-assembly skill that sourced public-domain footage and cut it into one continuous edit with clean narration.

In Google Search, Hydrogen is intended to deliver more helpful, intuitive responses for all users globally. Google said the model better understands a user's true intent, even for multi-part queries, hard-to-understand typos, or highly specific requests. Responses will be more visual, adaptive, and easy to scan at the speed users expect from Search. The company is also upgrading its query fan-out technique, using Hydrogen's advanced reasoning to connect users to a broader range of helpful links, including sites they might not otherwise have discovered.

Designed for safe, dependable execution at scale

Hydrogen carries improvements from Google's latest Gemini 4 safety and security stack to its broadest audience yet, combining built-in training with real-time surface safeguards to mitigate both misalignment and misuse. Google highlighted two focus areas:

Safe, aligned, and secure agentic workflows. While Gemini 3.8 Flash has strong alignment and safeguards, DeepMind said Hydrogen is a leap ahead. The model reduces overreach in agentic tasks by up to 81% and under-service by up to 79%, and cuts misaligned actions and task shortcutting across tool-use and coding evaluations by 12x. When faced with impossible or blocked tasks, it is up to 2x less likely to bypass permission boundaries and achieves a 96.7% honesty rate on Google's internal agent honesty evaluation. Building on the company's commitment to reasoning transparency, Hydrogen preserves 99.9% chain-of-thought legibility with zero deceptive reasoning edits across more than 200,000 audited turns. Against indirect prompt injection attacks, the model is more than 2.3x more resilient on the Gray Swan IPI benchmark, while significantly improving robustness across tool use, computer use, and coding agents.

Helpful, reliable everyday interactions. Across product surfaces, Google tuned Hydrogen to make everyday conversations more reliable and safe. Its updated safeguards are also more precise at distinguishing real biosecurity risks from legitimate medical or scientific questions, keeping false refusals on HealthBench Pro under 0.4%.

Further details are available in the Gemini 4 Hydrogen model card.

Gemini 4 Hydrogen, available to everyone today

Starting November 12, Hydrogen is rolling out globally across developer, enterprise, and consumer surfaces:

• For developers: Hydrogen can be built with through the Gemini API in AI Studio and the paid tier in Android Studio, tried in AI Studio, and tested in Google Antigravity by paid Google AI Pro and Ultra subscribers.
• For enterprises: The model can be deployed in Gemini for business via the API or model drop-down, in Google Antigravity, and in Gemini in Google Sheets for eligible business subscriptions.
• For everyone: It is available in the Gemini app model drop-down and in Google Search for all users, and in Gemini in Google Sheets starting with Google AI Pro and Ultra subscribers.

In the Gemini app, free users will get access to Gemini 4 Hydrogen via auto model selection when it is the appropriate model for a given prompt. That means free-tier access depends on auto selection classifying a prompt as a Hydrogen task, rather than on a manual opt-in.

About the author

Koray Kavukcuoglu is described by Google as one of the world's foremost experts in artificial intelligence. As SVP of Google DeepMind, he leads the development of the company's state-of-the-art generative AI models and their integration into Google products at scale. He previously served as VP of Research at Google DeepMind, where he founded the deep learning team; his team pioneered research breakthroughs including DQN (Deep Q-Network) and WaveNet.