On 12.11.2026, Google announced Gemini 4 Hydrogen, a model built for speed and efficiency in everyday coding and agentic workflows. While the company's frontier models, Gemini 4 Carbon and Gemini 4 Argon, target deep, long-horizon reasoning on the hardest problems, Hydrogen is meant for rapid execution across a wide range of agentic tasks. The release also brings Gemini 4 to all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app, and Google Search, plus paid Google AI Pro and Ultra subscribers in Google Antigravity.
Google says Hydrogen edges past Argon on the Artificial Analysis Intelligence Index at its default high thinking level, scoring 51 versus 50, and reaches 54 at its top xhigh setting, while often finishing tasks faster than other models in its class. Pricing is flat across the entire 1 million-token context window at $1.50 per million input tokens and $7.50 per million output tokens.
Multi-model workflows and Orchestrator
In real-world development, Google says teams lean on frontier models like Carbon and Argon for architectural judgment when designing systems or chasing subtle bugs, but still need a fast, disciplined model to write code, run tests, and verify changes. Hydrogen is designed to work on its own for everyday tasks or to act as the high-efficiency execution engine paired with Carbon and Argon.
To automate that pairing, Google is introducing Orchestrator in Google Antigravity. Instead of running a single model from start to finish, Orchestrator switches dynamically between models such as Carbon and Hydrogen within one session. On Terminal-Bench 4.0 it reached task success rates close to Carbon's while consuming 24% less quota on average and completing work 19% faster. For jobs that need parallel agents, Carbon can serve as the coordinator while Hydrogen powers worker and verifier agents that build and test changes at the same time in isolated workspaces, which Google says lets teams ship complex projects faster without inflating token costs.
Developer and enterprise workflows
Google positions Hydrogen as a major standalone upgrade for teams migrating from Gemini 3.8 Flash. The company says it completes tasks 2.5 times faster at its default high setting, more than triples the Terminal-Bench 4.1 score (57% versus 18.2%), cuts fabricated claims by more than five times, and adapts cleanly when a tool call fails or a search comes back empty. Google also says it matches or outperforms competing models in its class across core developer and enterprise work:
Iterative repository coding and debugging. Hydrogen handles issue resolution, debugging, rapid iteration, and codebase Q&A with focused, surgical edits that leave unrelated files untouched, and it beats similar-tier models on SWE-sweep, where it must find and fix bugs without breaking existing code. To move through large repositories quickly, it natively batches file-read and search calls in parallel instead of inspecting files one at a time. In early testing across complex graphs, SQL, and data-heavy enterprise workflows, Google reports that partners saw Hydrogen run 2x to 2.8x faster than 3.8 Flash while cutting token usage by 12% to 23%.
Open-ended agentic execution. On multi-step work such as deep research, data analysis, and broad codebase navigation, Hydrogen ranks fifth on Agent Arena, arena.ai's live benchmark. For deep domain reasoning it leads its class on HLE-Diamond at 58.8%, ranks first overall on the Vals Tax Agent Bench at 52.3%, ties for first on the Vals Legal Research Bench at 55.8%, and leads its class on AutomationBench-AA.
Multimodal knowledge work. Hydrogen pairs a native 1 million-token input window with up to 1 million output tokens to turn dense PDFs, spreadsheets, charts, audio, and recorded video into polished, cited reports and structured deliverables. Google says it outperforms other models on the new Multimodal Knowledge Agent (MKA) benchmark at 53.4% for end-to-end agentic work over complex visual files, and on Agents' Last Exam v1.1 at 71.6% for end-to-end professional workflows.
Web development. On speed-sensitive tasks like live UI prototyping, design-to-code iteration, and interactive app building, Google says Hydrogen produces more functional, feature-complete layouts in fewer prompts.
Four thinking levels
Hydrogen offers four thinking levels — low, medium, high (the default), and a new xhigh tier — spanning quality, latency, and cost in what Google describes as a smooth, predictable progression, so users only wait and pay for the reasoning depth a task needs. Across these levels, Google says Hydrogen matches or outperforms some competing models in its class while often finishing faster and using fewer output tokens.
The Gemini app and Google Search
For consumers, Google says Hydrogen stays on track and follows instructions more closely across multi-turn conversations. In the Gemini app it is presented as a more helpful everyday assistant that follows guidance better, calls the right tools, and reaches a reliable result with less back-and-forth.
For writing, research, and everyday work, Hydrogen reportedly sticks closely to the requested tone, format, and rules, so email drafts, study guides, rewrites, and documents come back ready to use on the first try, and it assembles clear tables and visual breakdowns for product comparisons or dense material without dropping key details. It also breaks a goal into steps, pulling context from uploads, Google apps such as Gmail and Google Drive, and Connected Apps like Airtable and Dropbox, and Google says it more often picks the right tools and skills when a task starts. For prototyping, it generates clean, working layouts in fewer prompts — Google gives the example of turning real-time global lightning strike data into an interactive 3D globe.
In Search, Google says Hydrogen delivers more helpful, intuitive responses for all users globally. It better understands the real intent behind multi-part questions, hard-to-read typos, and highly specific requests; returns more adaptive, visual, easy-to-scan responses at Search speed; and, through an upgraded query fan-out technique, surfaces a broader range of relevant links.
Safety and reliability
Hydrogen carries improvements from Google's latest Gemini 4 safety and security stack, combining built-in training with real-time surface safeguards to mitigate both misalignment and misuse. For a fast everyday model, Google highlights two areas.
On agentic safety, the company says Hydrogen goes well beyond 3.8 Flash: it reduces overreach in agentic tasks by up to 81% and under-service by up to 79%, cuts misaligned actions and task shortcutting by 12x across tool-use and coding evaluations, is up to 2x less likely to bypass permission boundaries on impossible or blocked tasks, and scores a 96.7% honesty rate on Google's internal agent honesty evaluation. Google also reports 99.9% chain-of-thought legibility with zero deceptive reasoning edits across more than 200,000 audited turns, and says Hydrogen is more than 2.3x more resilient against indirect prompt injection on the Gray Swan IPI benchmark, with improved robustness across tool use, computer use, and coding agents.
For everyday interactions, Google says its updated safeguards more precisely distinguish real biosecurity risks from legitimate medical or scientific questions, keeping false refusals on HealthBench Pro under 0.4%. More detail is available in the Gemini 4 Hydrogen model card.
Availability
The rollout begins 12.11.2026 across developer, enterprise, and consumer surfaces. Developers can use the Gemini API in AI Studio and the paid tier in Android Studio, test the model in AI Studio, and — as paid Google AI Pro or Ultra subscribers — use it in Google Antigravity. Enterprises can deploy it through Gemini for business via the API or the model dropdown, in Google Antigravity, and in Gemini in Google Sheets for eligible business subscriptions. Consumers can reach it in the Gemini app model dropdown and in Google Search for all users, with Gemini in Google Sheets starting with Google AI Pro and Ultra subscribers. In the Gemini app, free users get access to Hydrogen through auto model selection when it is the appropriate model for the prompt.