Google has launched Gemini 4 Hydrogen, a model built for rapid research, coding, and document work. Available across multiple Google products, it costs $1.50 per million input tokens and $7.50 per million output tokens. Speed and quota advantages in multi-model workflows are Google’s own claims, not independently verified results.
Hydrogen derives from Gemini 4 Carbon but is designed for fast execution of everyday tasks and agentic workflows. Google has made it available through the Gemini API, AI Studio, Gemini for Business, the Gemini app, and Google Search. In Antigravity, access is restricted to Google AI Pro and Ultra subscribers. In Google Sheets, Gemini functions powered by the model are tied to eligible subscriptions, starting from Pro and Ultra plans. The launch was announced on November 12, with Google’s product page updated the following day.
Pricing and Availability
API pricing sits at $1.50 per million input tokens and $7.50 per million output tokens, according to Google’s official documentation. Google states that the rate remains flat across the full one-million-token context window.
Availability is broad in principle, but different channels offer different features under different conditions. Access to Hydrogen in Antigravity depends on the user’s subscription tier, while Google indicates Search availability for all users. The company does not claim that every feature is live simultaneously in every country or account.
Google also positions Hydrogen as the engine for document and data tasks. The model card lists text, image, audio, and video inputs, with a context window of up to one million tokens. In Google Sheets, Google demonstrates the model building interactive templates from financial data and forecasting scenarios.
Orchestrator Coordinates Multiple Models
The most significant new capability for professional automation is Orchestrator, a feature inside Google Antigravity that can alternate between different models within a single session. In the workflow Google describes, a more powerful model like Carbon handles planning, while Hydrogen executes tasks or runs verification steps. Agents can work in parallel across isolated workspaces.
In tests reported by Google on Terminal-Bench 4.0, Orchestrator achieved success rates close to Carbon while consuming on average 24% less quota and completing tasks 19% faster. These are Google’s figures on a specific benchmark: the quota reduction figure does not translate into a universal 24% cut in monetary costs.
Google also reports that Hydrogen, compared with Gemini 3.8 Flash, completed certain tasks 2.5 times faster and scored 57% on Terminal-Bench 4.1, against 18.2% for the predecessor model. The company additionally claims a more than fivefold reduction in fabricated statements in its own comparison. These metrics depend on the tasks, settings, and testing methodology chosen by the vendor, so they don’t automatically describe performance across every enterprise use case.
Search Gets the New Model
Google states that Hydrogen has been integrated globally into Search. The company attributes to it better handling of multi-part questions, simultaneous research across several angles, and structured responses that include images, tables, and links. This is a change that reaches users who neither develop software nor build agents: the model enters a mainstream consumer product.
Hydrogen’s presence in Search does not mean every answer will be accurate, or that web browsing disappears behind automation. Google’s model card flags possible hallucinations, slowdowns, and timeouts, and notes that token consumption can increase, particularly at higher reasoning levels.
The Fast Model as Operational Backbone
Our read is that Google is shifting value away from single-model use toward distributing tasks across a fleet of models. Under this approach, the most capable model defines the plan while a faster model handles research, drafts materials, checks files, or runs repeated verification loops. The promise is to deploy expensive capacity only where it matters. Whether that translates into real savings depends on the specific workload and the number of steps involved.
For enterprises, API pricing, reasoning levels, and product access are all verifiable. The operational advantages Google cites are another matter: no independent data on the total costs of multi-agent workflows has been published, including error rates, human review steps, and potential task repetitions. Those are the numbers worth watching before committing budgets to agentic pipelines at scale.
