Introducing Gemini 4 Hydrogen

Google DeepMind introduced Gemini 4 Hydrogen, a model it positions as the fast, efficient counterpart to its frontier Gemini 4 Carbon and Gemini 4 Argon models. Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google, announced the launch, and the model is rolling out globally starting today across developer, enterprise and consumer surfaces. Google describes Carbon and Argon as handling deep, long-horizon reasoning on the hardest problems, while Hydrogen is aimed at rapid execution across agentic workflows. The release also marks the point where Gemini 4 reaches every tier of Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro and Ultra subscribers in Google Antigravity.
On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high thinking level versus 50 for Argon, and reaches 54 at a new xhigh setting. Pricing is flat across the full 1 million-token context window: $1.50 per million input tokens and $7.50 per million output tokens.
Hydrogen has four thinking levels: low, medium, high (the default) and the new xhigh. Google says they scale in a smooth, predictable progression so users pay and wait only for the reasoning depth a task needs. The model pairs a native 1 million-token input window with up to 1 million output tokens.
The positioning is explicit about multi-model work. Google argues that real software teams still rely on frontier models for architectural judgment and subtle debugging, but need a disciplined, fast model for writing code, running tests and verifying changes. Hydrogen can serve as a standalone everyday model or as the execution engine paired with Carbon and Argon.
To make that pairing automatic, Google is introducing Orchestrator in Google Antigravity, which switches between models such as Carbon and Hydrogen within a single session rather than using one model end to end. On Terminal-Bench 4.0, Orchestrator reached task success rates close to Carbon while using an average of 24% less quota and finishing 19% faster. For parallel-agent workflows, Google says Carbon can coordinate while Hydrogen powers worker and verifier agents working across isolated workspaces.
For teams moving off Gemini 3.8 Flash, Google reports Hydrogen 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 five times, and adapts better when a tool call fails or a search comes back empty. Partners testing complex graphs, SQL and data-heavy enterprise workflows saw 2x to 2.8x faster runs with 12% to 23% less token usage compared with 3.8 Flash.
Benchmark claims are broad. Hydrogen ranks #5 on Agent Arena, leads its class on HLE-Diamond at 58.8%, ranks #1 on the Vals Tax Agent Bench at 52.3%, ties for #1 on the Vals Legal Research Bench at 55.8%, and leads its class on AutomationBench-AA. It scores 53.4% on the new Multimodal Knowledge Agent benchmark and 71.6% on Agents' Last Exam v1.1. Google also highlights repository coding, where Hydrogen makes focused edits without rewriting unrelated files and natively batches parallel file-read and search calls, plus web development tasks such as live UI prototyping and design-to-code iteration.
Demonstrations include an on-the-fly 3D coding app called Geminisland, where a prompt adds voxel structures to a shared island players can explore in first person; a one-shot e-commerce storefront built with a new Gemini front end design skill in Antigravity, featuring a WebGL2 topographic shader and a 3D wireframe spec card; an animated pixel canvas built inside Google Sheets; and a multi-slide Google Slides presentation generated from a YouTube keynote and a design template, followed by an autonomous audit-and-fix loop. Google says Hydrogen generates outputs over 30% faster than previous models in Sheets canvas.
For consumers, Hydrogen is meant to follow instructions more closely across multi-turn conversations, call the right tools more often, and pull context from uploads, Gmail, Drive and Connected Apps such as Airtable and Dropbox. In Search, Google says Hydrogen better handles multi-part queries and typos, produces more visual responses, and upgrades its query fan-out technique to surface a wider range of links.
Safety claims are unusually specific. Google says Hydrogen reduces overreach in agentic tasks by up to 81% and under-service by up to 79%, cuts misaligned actions and task shortcutting by 12x, is up to 2x less likely to bypass permission boundaries on impossible or blocked tasks, and reaches a 96.7% honesty rate on an internal agent honesty evaluation. It preserves 99.9% chain-of-thought legibility with zero deceptive reasoning edits across more than 200,000 audited turns, and is more than 2.3x more resilient against indirect prompt injection on the Gray Swan IPI benchmark. False refusals on HealthBench Pro are held under 0.4%.
Developers can use Hydrogen through the Gemini API in AI Studio and the paid tier in Android Studio; paid AI Pro and Ultra subscribers can try it in Antigravity. Enterprises can deploy it in Gemini for business via the API or model dropdown, in Antigravity, and in Gemini in Google Sheets for eligible business subscriptions. Everyone can select it in the Gemini app model dropdown and in Google Search, with Sheets access starting with AI Pro and Ultra subscribers.
A likely trade-off is that the headline index win over Argon is narrow at default settings, so teams with hard reasoning workloads will probably still route those to Carbon or Argon and use Hydrogen for execution throughput. The flat context pricing is notable: many providers charge more for long prompts, so a single rate across 1 million tokens simplifies cost modeling for large-repository or document-heavy agents.
Why it matters: Developers and product teams get a cheaper, faster Gemini 4 tier with the same context window as the frontier models, plus an Antigravity feature that spreads a session across models to save quota and time. Teams migrating from Gemini 3.8 Flash have concrete speed and benchmark numbers to weigh, and the safety metrics for agent overreach and prompt injection are the kind of numbers enterprises will ask about before deploying agents. Whether Hydrogen's narrow edge over Argon holds across real workloads is likely to be tested quickly by early adopters.
Based on reporting from the original publisher. Visit the source for full context and later updates.