What is Gemini 4 Hydrogen and who is it for

Google DeepMind introduced Gemini 4 Hydrogen on November 12, 2026, as a new model for everyday coding, agentic workflows, and search.

Hydrogen is the successor to Gemini 3.8 Flash, dropping the Flash naming and bringing Gemini 4-class intelligence to a broader audience.

In a blog post, Koray Kavukcuoglu, SVP at Google DeepMind and chief AI architect, said the model brings cutting-edge intelligence to everyone with speed and efficiency for everyday use.

While Gemini 4 Carbon and Gemini 4 Argon handle deep, long-horizon reasoning on the hardest problems, Hydrogen is built for rapid execution across diverse agentic workflows.

The company positions it as a model that can stand on its own for everyday work or pair with Carbon and Argon as the high-efficiency execution engine in multi-model setups.

Hydrogen is available across every Google AI tier: Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro and Ultra subscribers in Google Antigravity.

Free users in the Gemini app get access through auto model selection when Hydrogen is the appropriate model for their prompt, without needing a paid subscription.

The wide availability marks a shift from previous Flash models that were primarily available through APIs and the Gemini app only.

How fast and capable is Hydrogen compared to the previous Flash model

Gemini 4 Hydrogen finishes tasks 2.5 times faster at its high default setting compared to Gemini 3.8 Flash. Its Terminal-Bench 4.1 score more than triples to 57%, up from 18.2%.

The model cuts fabricated claims by more than 5 times and adapts cleanly when a tool call fails or a search returns empty, a common failure point in agentic workflows.

In early testing across complex graphs, SQL, and data-heavy enterprise workflows, partners saw Hydrogen run 2 to 2.8 times faster while reducing token usage by 12% to 23%, compared to 3.8 Flash.

Hydrogen also 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 12 times.

When faced with impossible or blocked tasks, it is up to 2 times less likely to bypass permission boundaries and achieves a 96.7% honesty rate on Googles internal agent honesty evaluation.

On multi-step tasks like deep research, data analysis, and broad codebase navigation, Hydrogen ranks fifth on Agent Arena, the live benchmark from arena.ai.

It outperforms other models on the Multimodal Knowledge Agent benchmark with 53.4% for end-to-end agentic work over complex visual files, and scores 71.6% on Agents Last Exam v1.1 for professional workflows.

Hydrogen beats Argon on intelligence benchmarks at default settings

Hydrogen surpasses Gemini 4 Argon on the Artificial Analysis Intelligence Index at its default high thinking level, scoring 51 versus 50, and reaching 54 at its highest thinking setting, while often completing tasks faster than models in its class.

The AA Index measures intelligence across reasoning, coding, math, and language understanding, making the crossover between a efficiency-focused model and a frontier model notable.

For deep domain reasoning and analysis, Hydrogen leads its class on HLE-Diamond at 58.8%, a benchmark that tests graduate-level expertise across hard sciences and humanities.

It ranks first overall on the Vals Tax Agent Bench with 52.3% and ties for first on the Vals Legal Research Bench with 55.8%, both benchmarks that require structured reasoning through complex tax and legal documents.

The model also leads its class on AutomationBench-AA, which measures how reliably models complete enterprise automation tasks without human guidance.

On multimodal agentic work, Hydrogen outperforms competing models on the Multimodal Knowledge Agent benchmark with 53.4% for end-to-end knowledge work over complex visual files like charts, diagrams, and scanned documents.

On Agents Last Exam v1.1, it scores 71.6% for professional workflows spanning data analysis, code generation, and document synthesis.

Hydrogen powers Google Search AI Mode and the Gemini app

Hydrogen is the default model for Google Search AI Mode, replacing Gemini 3.8 Flash as the engine behind conversational search responses. Google says the upgrade significantly improves understanding of user intent, including multi-part queries, hard-to-understand typos, and highly specific requests.

Search responses are now more visual and feature a wider range of relevant links rather than only AI-generated summaries. Hydrogen is also used for the most difficult queries in AI Overviews.

In the Gemini app, the model is better at following and sticking to tone, format, and rules. According to Google, email drafts, study guides, rewrites, and documents come back ready to use on the first try.

Hydrogen also more frequently calls the right tools and skills, including Connected Apps integrations.

The app model picker now shows four tiers: Quick answers powered by 3.5 Flash-Lite, Everyday help powered by Hydrogen, Detailed research powered by 3.1 Pro, and Code and analysis powered by Argon.

Free users get access to Hydrogen through auto model selection when it is the appropriate model for the prompt. Paid users on AI Pro and Ultra plans can pick Hydrogen manually alongside Carbon, Argon, and the new Orchestrator mode.

What is Orchestrator and how does it work with Hydrogen

Alongside Hydrogen, Google introduced Orchestrator in Antigravity. Rather than using a single model from start to finish, Orchestrator dynamically switches between models like Carbon and Hydrogen within the same session.

On Terminal-Bench 4.0, Orchestrator reached near-Carbon task success rates while consuming 24% less quota and completing 19% faster.

For workflows that require parallel agents, Carbon can act as the coordinator while Hydrogen powers worker and verifier agents to build and test changes simultaneously across isolated workspaces.

On Terminal-Bench 4.0, Orchestrator reached near-Carbon task success rates while consuming 24% less quota and completing 19% faster. Google says this helps teams ship complex projects faster without driving up token costs.

Pricing is flat across the entire 1 million-token context window, a change from many models that charge different rates for short and long contexts.

This means developers pay the same per-token rate whether their prompt uses 100 tokens or fills the full one million.

For comparison, Gemini 4 Carbon costs $4 per million input tokens and $20 per million output tokens, making Hydrogen significantly cheaper for high-volume agentic workloads.

Gemini 4 Hydrogen pricing
MetricPrice
Input tokens$1.50 per million tokens
Output tokens$7.50 per million tokens
Context window1 million tokens

Common questions

Is Gemini 4 Hydrogen free to use?

Free users in the Gemini app get access to Hydrogen through the auto model selection when Hydrogen is the appropriate model for the prompt.

The model is also available to all tiers of Google AI Studio, the Gemini API, Gemini for business, and paid Pro and Ultra subscribers in Google Antigravity.

How does Gemini 4 Hydrogen compare to Gemini 4 Carbon?

Carbon is the frontier model for deep, long-horizon reasoning on the hardest problems. Hydrogen is built for rapid execution across diverse agentic workflows.

Hydrogen surpasses Argon on the Artificial Analysis Intelligence Index at default settings but does not match Carbon's peak reasoning depth. The two models can work together through Orchestrator.

What happened to the Gemini Flash naming?

Google dropped the Flash branding with this release. Gemini 4 Hydrogen replaces what would have been the Flash tier, bringing Gemini 4-class intelligence to everyday tasks rather than keeping a separate naming convention for its efficient model.

Sources & further reading

  1. Google DeepMind - Introducing Gemini 4 Hydrogen for Agentic Workflows ↗
  2. Google Blog - Introducing Gemini 4 Hydrogen for Agentic Workflows ↗
  3. 9to5Google - Google announces Gemini 4 Hydrogen with significant leaps in speed, intelligence ↗
  4. The Verge - Google's new Gemini model is better at understanding complicated questions ↗
How this story was made

Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. The cover is an AI-generated editorial illustration, not a photograph or evidence of an actual event.

Our editorial standards
Back to all stories