Google Launches Gemini 4 Hydrogen, Its Fast Everyday Model for Search, Coding, and Agents
Google Launches Gemini 4 Hydrogen, Its Fast Everyday Model for Search, Coding, and Agents
Google DeepMind introduced Gemini 4 Hydrogen on November 12, 2026, a fast, lower-cost model built for everyday coding and agentic work. It is the third model in the Gemini 4 family, following Argon in September and Carbon in October, and it is the first of the three to reach everyone on day one. Hydrogen replaces Gemini 3.8 Flash across Google’s products, and it is rolling out globally in the Gemini app, Google Search, Google AI Studio, the Gemini API, Gemini for business, and Google Antigravity for paid Pro and Ultra subscribers.
Hydrogen is positioned as the execution tier of the Gemini 4 family. Carbon and Argon remain the frontier models for deep, long-horizon reasoning on hard problems. Hydrogen is built for rapid execution: writing code, running tests, verifying changes, and carrying out multi-step tasks inside agents. Google describes it as a model that can stand on its own for everyday work or pair with Carbon and Argon in multi-model workflows, where a frontier model plans and Hydrogen executes.
A day-one rollout across Search, the app, and the API
Unlike its siblings, Hydrogen skipped the staged rollout. Argon first went to vetted defenders in Google’s Fairwind cybersecurity program and reached general availability two weeks later. Carbon followed on October 28 and only opened to paying developers on November 11. Hydrogen arrived the same day for developers through the Gemini API and AI Studio on all tiers, for businesses through Gemini for business and in Google Sheets, and for consumers in the Gemini app and in Search for all users.
In Search, Google calls Hydrogen its biggest model upgrade yet. The company says Hydrogen better understands a question asked in natural language, including questions with multiple parts, hard-to-understand typos, or highly specific requests. It also promises more visual, easy-to-scan responses and a wider range of relevant links, because Hydrogen upgrades the query fan-out technique Search uses to connect a question to helpful sites. In the Gemini app, Google says Hydrogen follows instructions more closely across multi-turn conversations, calls the right tools more often, and produces drafts and documents that need less back-and-forth.
Speed, price, and four thinking levels
Google prices Hydrogen flat across its entire 1 million-token context window at $1.50 per million input tokens and $7.50 per million output tokens. The model pairs 1 million input tokens with up to 1 million output tokens, which Google says lets it synthesize dense PDFs, spreadsheets, charts, audio, and recorded video into cited reports and structured deliverables.
Hydrogen offers four thinking levels: low, medium, high as the default, and a new xhigh tier. The levels let users trade reasoning depth against latency and cost, so a quick question can run at low while a harder task gets the xhigh setting. At its default high setting, Google reports Hydrogen scores 51 on the Artificial Analysis Intelligence Index, one point above Argon at 50, and reaches 54 at xhigh. Google also says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at the default setting.
On coding benchmarks, Google reports Hydrogen more than tripled its predecessor’s Terminal-Bench 4.1 score, to 57 percent from 18.2 percent, and cut fabricated claims by more than five times. In early testing with partners, Google says Hydrogen ran 2 to 2.8 times faster than 3.8 Flash while using 12 to 23 percent fewer tokens. On Agent Arena, a live benchmark of multi-step agent tasks, Hydrogen ranks fifth overall. All of these figures come from Google, and independent testing has not yet confirmed them.
Orchestrator pairs the models automatically
Alongside Hydrogen, Google introduced Orchestrator in Google Antigravity, a feature that switches between models such as Carbon and Hydrogen within a single session. The idea is to use the frontier model where deep judgment matters and hand execution to Hydrogen where speed and cost matter. On Terminal-Bench 4.0, Google reports Orchestrator reached near-Carbon task success rates while consuming 24 percent less quota and finishing 19 percent faster on average.
Safety work aimed at agent reliability
Because Hydrogen is built for agents that act on a user’s behalf, Google published agent safety results from its internal testing. The company says Hydrogen reduces overreach in agentic tasks by up to 81 percent and cuts misaligned actions and task shortcutting across tool-use and coding evaluations by 12 times. When faced with an impossible or blocked task, Google says it is up to two times less likely to bypass permission boundaries, and it reports a 96.7 percent honesty rate on an internal agent honesty evaluation. Against indirect prompt injection attacks, Hydrogen is more than 2.3 times more resilient on the Gray Swan IPI benchmark than its predecessor, according to Google.
Why it matters
Hydrogen matters less as a new capability ceiling than as a shift in where Google’s newest intelligence actually runs. Fast, inexpensive models now carry most everyday AI traffic: they answer search queries, fill spreadsheets, and run inside agents that make many calls per task. By putting a Gemini 4 model into Search for all users on day one, Google is testing its latest generation in front of its largest audience at once. For developers, the practical change is simpler: the migration path from 3.8 Flash now points to a single model that is faster, scores higher on the benchmarks Google cites, and plugs into the same API and tooling. The open question is whether Google’s speed and reliability claims hold up outside its own testing.
