Google DeepMind ships Gemini 4 Hydrogen, a fast agentic model

Google DeepMind has introduced Gemini 4 Hydrogen, a model in the Gemini 4 family aimed at speed and efficiency in everyday coding and agentic workflows. The post positions it below the family's frontier models, Gemini 4 Carbon and Gemini 4 Argon, which it says excel at deep, long-horizon reasoning on the hardest problems. Hydrogen, by contrast, is described as built for rapid execution across diverse agentic workflows.
On price and benchmarks, the post says Hydrogen surpasses Argon on the Artificial Analysis Intelligence Index at its default high thinking level, 51 against 50, and reaches 54 at its highest xhigh setting, while often completing tasks faster than models in its class. It is priced flat across its whole 1 million-token context window: $1.50 per million input tokens and $7.50 per million output tokens. It pairs native 1 million-token input with up to 1 million output tokens.
A second announcement is Orchestrator in Google Antigravity. Instead of using one model from start to finish, Orchestrator switches dynamically between models such as Carbon and Hydrogen within the same session. On Terminal-Bench 4.0 it reached near-Carbon task success rates while, on average, consuming 24% less quota and completing 19% faster. For parallel-agent workflows, Carbon can act as coordinator while Hydrogen powers worker and verifier agents that build and test changes at the same time across isolated workspaces. Hydrogen is also meant to stand on its own, or to serve as the high-efficiency execution engine next to Carbon and Argon.
For teams migrating from Gemini 3.8 Flash, the post calls Hydrogen a major standalone upgrade. It says Hydrogen finishes tasks 2.5 times faster at its high default setting, more than triples the Terminal-Bench 4.1 score (57% against 18.2%), cuts fabricated claims by more than 5x, and adapts cleanly when a tool call fails or a search returns empty. The post says it matches or outperforms competing models in its class across core developer and enterprise workflows, without naming those models.
The post groups the evidence by workload. In repository coding and debugging, Hydrogen is said to make focused, surgical edits without rewriting unrelated files, to outperform similar-tier models on SWE-sweep, and to batch file-read and search calls in parallel. In early testing on complex graphs, SQL and data-heavy enterprise workflows, unnamed partners saw it run 2x to 2.8x faster with 12% to 23% lower token usage than 3.8 Flash. For open-ended agentic work it ranks #5 on Agent Arena, arena.ai's live benchmark. It is said to lead its class on HLE-Diamond (58.8%), rank #1 overall on Vals Tax Agent Bench (52.3%), tie for #1 on the Vals Legal Research Bench (55.8%) and lead its class on AutomationBench-AA. For multimodal knowledge work it is said to outperform other models on the new Multimodal Knowledge Agent (MKA) benchmark (53.4%) and on Agents' Last Exam v1.1 (71.6%). For web development, it is said to produce more functional, feature-complete layouts in fewer prompts.
Hydrogen has four thinking levels: low, medium, high (the default) and a new xhigh tier. The post says they scale smoothly so users wait and pay only for the reasoning depth a task needs, and that across levels Hydrogen matches or outperforms some of the competing models in its class while often finishing faster and using fewer output tokens.
For consumers, the post says Hydrogen follows instructions more closely over multi-turn conversations. In the Gemini app it is said to stick to tone, format and rules in drafts and documents, pull context from uploads, Gmail, Google Drive and Connected Apps such as Airtable and Dropbox, and call the right tools more often. It also generates working prototypes in fewer prompts, for example an interactive 3D globe of real-time lightning strike data. In Search, Hydrogen is said to better understand queries with several parts or typos, give more visual and easy-to-scan answers, and power an upgraded query fan-out technique that connects users to a broader range of links, including sites they might not have found before.
On safety, the post says Hydrogen brings improvements from the latest Gemini 4 safety and security stack, combining built-in training with real-time surface safeguards. Against 3.8 Flash, it reports reducing overreach in agentic tasks by up to 81% and under-service by up to 79%, and cutting misaligned actions and task shortcutting across tool-use and coding evaluations by 12x. On impossible or blocked tasks it is up to 2x less likely to bypass permission boundaries and reaches a 94% honesty rate on Google's 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 on the Gray Swan IPI benchmark against indirect prompt injection. False refusals on HealthBench Pro stay under 0.4%. The post points to a model card and a Safety and Security Report for details.
Availability: starting today Hydrogen is rolling out globally. Developers get it through the Gemini API in AI Studio and Android Studio, and paid Google AI Pro or Ultra subscribers can try it in Google Antigravity. Enterprises can deploy it in Gemini for business via the API or model drop down, in Google Antigravity, and in Gemini in Google Sheets for eligible business subscriptions. Everyone can use it in the Gemini app model drop down and in Google Search, and in Gemini in Google Sheets starting with Google AI Pro and Ultra subscribers.
Key facts
- Gemini 4 Hydrogen is priced flat across its 1 million-token context window at $1.50 per million input tokens and $7.50 per million output tokens, with up to 1 million output tokens.
- Google says Hydrogen scores 51 against Argon's 50 on the Artificial Analysis Intelligence Index at its default high thinking level, and 54 at the new xhigh setting.
- Against Gemini 3.8 Flash, Hydrogen finishes tasks 2.5 times faster, scores 57% vs. 18.2% on Terminal-Bench 4.1 and cuts fabricated claims by more than 5x.
- Orchestrator in Google Antigravity switches between models such as Carbon and Hydrogen in one session; on Terminal-Bench 4.0 it reached near-Carbon success while using 24% less quota and finishing 19% faster, on average.
- Hydrogen is rolling out globally from the announcement day across AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search and Google Antigravity for paid Pro and Ultra subscribers.
Why it matters
Hydrogen is the fast, everyday tier of the Gemini 4 family, and it is being put in front of the broadest audience at once: developers, enterprises and consumers, including every user of Google Search. Google claims it beats Argon on one third-party index at its default setting, which would make speed less of a trade against capability. The Orchestrator feature points at a workflow where a frontier model plans and a cheaper model executes and verifies, with the model switching handled automatically.
Who it affects
Developers building on the Gemini API and AI Studio, and teams still on Gemini 3.8 Flash, which the post singles out as the migration path. Enterprise customers of Gemini for business and Gemini in Google Sheets. Paid Google AI Pro and Ultra subscribers using Google Antigravity. Everyday users of the Gemini app and Google Search, where Hydrogen also changes how queries are interpreted and how links are fanned out.
How to use it
Developers can build with Hydrogen through the Gemini API in AI Studio and Android Studio. Enterprises can use it in Gemini for business via the API or model drop down, in Google Antigravity, and in Gemini in Google Sheets for eligible business subscriptions. Consumers can pick it in the Gemini app model drop down and get it in Google Search; in Gemini in Google Sheets it starts with Pro and Ultra subscribers. Google Antigravity access requires a paid Google AI Pro or Ultra subscription. API pricing is $1.50 per million input tokens and $7.50 per million output tokens, flat across the 1 million-token window. Four thinking levels (low, medium, high as the default, and xhigh) let users trade reasoning depth against latency and cost. In parallel-agent setups, the post suggests Carbon as coordinator with Hydrogen as worker and verifier agents.
How solid is it
This is a first-party announcement from Google DeepMind, and the rankings and leads are the vendor's own claims, not independent results. The competing models in "its class" are not named. The post says Hydrogen matches or outperforms competing models across core workflows, but across thinking levels only some of them. The partners behind the 2x to 2.8x speedups are not named. The Orchestrator result is on Terminal-Bench 4.0, while the Hydrogen vs 3.8 Flash result is on Terminal-Bench 4.1, and the post does not equate them. The Artificial Analysis figures (51, 50, 54) come with no stated scale or maximum. The post does not say how the internal honesty evaluation or the audited-turns sample was built.
Risks and caveats
No pricing, availability dates or plan limits for Orchestrator are given. No prices, context windows or benchmark scores for Carbon and Argon are given, and no Gemini 3.8 Flash price is given, so a cost comparison with Hydrogen cannot be made from the post. The post gives no timeline for when each surface will have Hydrogen beyond "starting today" and "rolling out globally". The faster-task claim is hedged in the post itself: Hydrogen is said to often, not always, complete tasks faster than models in its class. Safety figures such as the 94% honesty rate come from Google's own evaluations; the post refers readers to the model card and the Safety and Security Report for details.
“Hydrogen is built for rapid execution across diverse agentic workflows.”
— Google DeepMind, announcement post