Google ships Gemini 4 Hydrogen, a fast agentic model, and Antigravity Orchestrator

Google announced Gemini 4 Hydrogen, a model built for rapid execution in coding and agentic workflows. It sits beside the family's frontier models, Gemini 4 Carbon and Gemini 4 Argon, which Google says excel at deep, long-horizon reasoning on the hardest problems. Hydrogen is rolling out the same day across all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, and to paid Google AI Pro and Ultra subscribers in Google Antigravity.
Google's headline claim is that Hydrogen surpasses Argon on the Artificial Analysis Intelligence Index at its default high thinking level, 51 against 50, and reaches 54 at its new xhigh setting, while often finishing tasks faster than models in its class. Pricing is flat across the whole 1 million-token context window: $1.50 per million input tokens and $7.50 per million output tokens. Hydrogen takes native input of 1 million tokens and can produce up to 1 million output tokens. It has four thinking levels: low, medium, high (the default) and xhigh.
The 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 in the same session. On Terminal-Bench 4.0, Google says it reached near-Carbon task success rates while, on average, using 24% less quota and finishing 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 in isolated workspaces.
Against Gemini 3.8 Flash, which Google calls the baseline for teams migrating, Hydrogen finishes tasks 2.5 times faster at its high default setting and more than triples the Terminal-Bench 4.1 score, 57% against 18.2%. Google also says it cuts fabricated claims by more than 5x and recovers better when a tool call fails or a search returns nothing. In early testing by partners on complex graphs, SQL and data-heavy enterprise workflows, Hydrogen ran 2x to 2.8x faster than 3.8 Flash with 12% to 23% fewer tokens.
Google lists results by workload. For repository coding and debugging, it says Hydrogen makes focused edits without rewriting unrelated files, outperforms similar-tier models on SWE-sweep, and batches file-read and search calls in parallel. For open-ended agentic work it ranks #5 on Agent Arena, arena.ai's live benchmark. 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. For multimodal knowledge work, which covers PDFs, spreadsheets, charts, audio and recorded video, Google says it outperforms other models on the new Multimodal Knowledge Agent benchmark (53.4%) and on Agents' Last Exam v1.1 (71.6%). For web development, it says Hydrogen produces more functional layouts in fewer prompts.
In the Gemini app, Google says Hydrogen follows tone, format and rules more closely, calls the right tools more often, and pulls context from uploads, Gmail, Google Drive and Connected Apps such as Airtable and Dropbox. In Search, Hydrogen is described as a significant upgrade for all users globally: better reading of multi-part or typo-ridden queries, more visual responses, and an upgraded query fan-out technique that surfaces a wider range of links.
The safety section compares Hydrogen with 3.8 Flash using Google's own evaluations. Google says it 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 tasks are impossible or blocked, it is up to 2x less likely to bypass permission boundaries, with a 96.7% 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, is more than 2.3x more resilient on the Gray Swan indirect prompt injection benchmark, and keeps false refusals on HealthBench Pro under 0.4%. A model card carries further detail.
Key facts
- Gemini 4 Hydrogen is the execution-focused model of the Gemini 4 family; Google says it scores 51 against Argon's 50 on the Artificial Analysis Intelligence Index at high thinking, and 54 at xhigh.
- Price is flat at $1.50 per million input tokens and $7.50 per million output tokens across the full 1 million-token context window.
- Versus Gemini 3.8 Flash, Google reports 2.5 times faster task completion, Terminal-Bench 4.1 up from 18.2% to 57%, and fabricated claims cut by more than 5x.
- Antigravity's new Orchestrator switches between models such as Carbon and Hydrogen in one session; on Terminal-Bench 4.0 it reached near-Carbon success with 24% less quota and 19% faster completion.
- Rollout started the same day across the Gemini API, AI Studio, Gemini for business, the Gemini app and Search, and to paid Google AI Pro and Ultra subscribers in Antigravity.
Why it matters
Google is splitting the Gemini 4 family by job: Carbon and Argon for deep, long-horizon reasoning, Hydrogen for fast execution such as writing code, running tests and verifying changes. The Orchestrator in Antigravity turns that split into a product feature by choosing models within a single session. Google also claims Hydrogen beats Argon on one aggregate index (51 against 50) at a stated price of $1.50 per million input tokens and $7.50 per million output tokens, and puts it into Search for all users globally.
Who it affects
Developers using the Gemini API, AI Studio or Android Studio; enterprises using Gemini for business or Gemini in Google Sheets on eligible subscriptions; paid Google AI Pro and Ultra subscribers in Antigravity; and ordinary users of the Gemini app and Google Search. Teams still on Gemini 3.8 Flash are the group Google addresses most directly, calling Hydrogen a major standalone upgrade for them.
How to use it
Build with the Gemini API in AI Studio and Android Studio. Paid Google AI Pro or Ultra subscribers can try it in Google Antigravity, where Orchestrator can pair it with Carbon. Enterprises can deploy it in Gemini for business via the API or the model drop-down, in Antigravity, and in Gemini in Google Sheets for eligible business subscriptions. Everyone else can pick it in the Gemini app model drop-down or get it in Search, and in Gemini in Google Sheets starting with Pro and Ultra subscribers. API pricing is $1.50 per million input tokens and $7.50 per million output tokens, flat across the 1 million-token window. Thinking levels run low, medium, high (default) and xhigh, so you pay for only as much reasoning depth as a task needs.
How solid is it
Every number comes from Google's own announcement. The Artificial Analysis figures are quoted by Google, and the post does not say who ran them or how. The safety results are Google's internal evaluations measured against 3.8 Flash. The competing models in its class are not named. The post does not say what separates Terminal-Bench 4.0 from 4.1, and gives no score for SWE-sweep or AutomationBench-AA. A model card is linked for more detail.
Risks and caveats
Several claims are maximums or lack a baseline: overreach down by "up to" 81%, under-service by "up to" 79%, and "up to 2x" less likely to bypass permission boundaries. The 2.3x Gray Swan resilience figure does not name its baseline in that sentence. Some benchmarks are new (the Multimodal Knowledge Agent benchmark) or unexplained, so they are hard to judge. The post gives no prices for Carbon or Argon and no pricing or quota details for the Gemini app and Search. Orchestrator's results are reported on a single benchmark, Terminal-Bench 4.0, and as averages.