Google Launches Gemini 4 Hydrogen, a Faster AI Model Built for Coding and Agentic Work

Google has introduced Gemini 4 Hydrogen, a new artificial intelligence model built for speed. Announced on November 12, 2026, Hydrogen is aimed at everyday coding and the multi-step, tool-using workflows that have come to be known as agentic AI, and it is rolling out globally from today across Google's developer, business and consumer products.

The launch fills a gap in Google's Gemini 4 family. While the frontier models Gemini 4 Carbon and Gemini 4 Argon are designed for deep, long-horizon reasoning on the hardest problems, Hydrogen is the execution engine: the model that writes the code, runs the tests, checks the results and keeps a project moving. Google says it can work on its own, or pair with Carbon and Argon in multi-model workflows where the heavier models handle planning and judgment while Hydrogen handles the doing.

Where Hydrogen is available

Google is bringing Gemini 4 to everyone with this release. Hydrogen is available across all tiers of Google AI Studio and the Gemini API, in Gemini for business, in the Gemini app and in Google Search, and for paid Google AI Pro and Ultra subscribers in Google Antigravity, Google's agentic development platform. In the Gemini app, free users get access through automatic model selection when Google judges Hydrogen the right model for a prompt.

The name will be familiar to close watchers of Google's AI releases. Hydrogen had been spotted in tester reports and benchmark sightings for weeks before the announcement, but Google had not confirmed the model until now.

Speed, thinking levels and price

Hydrogen is built around a simple trade-off: give developers control over how much reasoning a task gets, and charge the same rate whatever depth they choose. The model offers four thinking levels, low, medium, high and a new xhigh tier, with high as the default. Google says the levels scale in a smooth, predictable progression, so users only wait and pay for the reasoning depth a task actually needs.

On Google's own benchmarks, Hydrogen edges out Argon on the Artificial Analysis Intelligence Index at its default high setting, scoring 51 against Argon's 50, and reaches 54 at the xhigh setting, while often finishing tasks faster than models in its class. Against the model it effectively replaces, Gemini 3.8 Flash, Google reports that Hydrogen finishes tasks 2.5 times faster at its high default setting, more than triples Terminal-Bench 4.1 scores to 57 percent against 18.2 percent, and cuts fabricated claims by more than five times.

Pricing is flat across the model's entire one-million-token context window: 1.50 dollars per million input tokens and 7.50 dollars per million output tokens. All of these figures are Google's own evaluations, and developers will want to test them against their own workloads before drawing conclusions about real-world cost or speed.

Orchestrator: switching models inside one session

Alongside Hydrogen, Google is introducing Orchestrator in Google Antigravity, a system that dynamically switches between models such as Carbon and Hydrogen within a single development session instead of using one model from start to finish. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon task success rates while consuming 24 percent less quota on average and completing tasks 19 percent faster.

The same idea extends to parallel agents. Carbon can act as the coordinator while Hydrogen powers worker and verifier agents that build and test changes simultaneously across isolated workspaces, a setup aimed at teams shipping complex projects without driving up token costs.

What Hydrogen is good at, according to Google

For software teams, Google positions Hydrogen as a major standalone upgrade for work migrating from Gemini 3.8 Flash. It handles issue resolution, debugging, rapid iteration and codebase questions with focused, surgical edits, and in early partner testing across complex graphs, SQL and data-heavy enterprise workflows, it ran two to 2.8 times faster while reducing token usage by 12 to 23 percent compared with 3.8 Flash.

On open-ended agentic work such as deep research, data analysis and broad codebase navigation, Hydrogen ranks fifth on Agent Arena, a live benchmark, and leads its class on several specialist tests, including HLE-Diamond at 58.8 percent, the Vals Tax Agent Bench at 52.3 percent and the Vals Legal Research Bench at 55.8 percent. For multimodal knowledge work, it pairs a native one-million-token input window with up to one million output tokens, synthesising dense PDFs, spreadsheets, charts, audio and recorded video into cited reports. It also scores 53.4 percent on the new Multimodal Knowledge Agent benchmark and 71.6 percent on Agents' Last Exam v1.1.

Everyday help in the Gemini app and Search

For consumers, Hydrogen is meant to make the Gemini app a more reliable everyday assistant, following instructions more closely across multi-turn conversations, calling the right tools and reaching a usable output with less back-and-forth. It can pull context from uploads, Gmail and Google Drive, and from connected apps such as Airtable and Dropbox, and it generates working layouts for prototyping in fewer prompts.

Google Search is also getting a significant upgrade from Hydrogen. Google says Search will understand questions asked in a natural flow of thought, even with multiple parts, typos or highly specific requests, return more visual and easy-to-scan responses, and use improved query fan-out to connect users with a broader range of relevant links.

Safety work behind the faster model

Google says Hydrogen carries improvements from its latest Gemini 4 safety and security stack to its broadest audience yet. For agentic work, the company reports that Hydrogen reduces overreach in agentic tasks by up to 81 percent and under-service by up to 79 percent, and cuts misaligned actions and task shortcutting across tool-use and coding evaluations by twelve times. When faced with impossible or blocked tasks, it is up to two times less likely to bypass permission boundaries, and it records a 96.7 percent honesty rate on Google's internal agent honesty evaluation.

The model also preserves 99.9 percent chain-of-thought legibility with no deceptive reasoning edits across more than 200,000 audited turns, according to Google, and is more than 2.3 times more resilient against indirect prompt injection attacks on the Gray Swan IPI benchmark. False refusals on HealthBench Pro stay under 0.4 percent. As with the performance figures, these are Google's own measurements, and the company's model card acknowledges that hallucinations, occasional timeouts and some unjustified refusals can still occur.

Why it matters

Hydrogen arrives days after the broader availability of Gemini 4 Carbon, and together the two releases sketch out Google's strategy for the Gemini 4 generation: expensive reasoning used sparingly, fast execution used everywhere. If the benchmarks hold up in real use, the model could make agentic coding tools cheaper and quicker for everyday developers, and bring the newest Gemini generation into Search and the Gemini app for free users at the same time. The test, as ever with a launch built on company-reported numbers, will be what developers and users find when they run it on their own work.

Sources

• Google, "Introducing Gemini 4 Hydrogen for Agentic Workflows", The Keyword, November 12, 2026 — https://blog.google/intl/en-in/products/introducing-gemini-4-hydrogen/
• Pulse2, "Google: Gemini 4 Hydrogen Launches With Faster Agentic AI, 1 Million-Token Context Window, And New Multi-Model Orchestration", November 13, 2026 — https://pulse2.com/google-gemini-4-hydrogen-launches-with-faster-agentic-ai-1-million-token-context-window-and-new-multi-model-orchestration/
• Yowox, "Gemini 4 Hydrogen adds a faster execution tier", November 13, 2026 — https://yowox.com/posts/gemini-4-hydrogen-adds-faster-execution-tier/