Google Gemini 4 Hydrogen replaces 3.8 Flash and powers Search

On 12 November 2026, Google introduced Gemini 4 Hydrogen, a fast and inexpensive language model that replaces Gemini 3.8 Flash across the company's products. According to Google DeepMind, the model is built for everyday coding and for agentic workflows, and it reaches every Google Search user worldwide from day one. Ars Technica notes that, unlike the earlier Carbon and Argon releases, there is no staged, security-first rollout this time.

What is Gemini 4 Hydrogen and where does it fit?

Gemini 4 is a family of three models. According to Google, Carbon and Argon are the frontier models, meant for deep, long-horizon reasoning on the hardest problems, while Hydrogen is built for rapid execution. Ars Technica observes that the elemental names are less descriptive than the old Pro and Flash labels, but that Hydrogen's role is clear: it takes over from the faster, cheaper Flash models that most people in Google's ecosystem use.

The term agentic workflow, as TechCrunch explains it, describes tasks in which an AI carries out several steps on its own. Google says Hydrogen can handle everyday work such as writing code, running tests and verifying changes by itself, or act as the high-efficiency execution engine next to Carbon and Argon in multi-model setups.

According to Google DeepMind, the model is available in all tiers of Google AI Studio, in the Gemini API, in Gemini for business, in the Gemini app and in Google Search. In Google Antigravity, its development platform, it is open to paid AI Pro and Ultra subscribers.

How much faster and smarter is it than Gemini 3.8 Flash?

• Speed: tasks finish 2.5 times faster at the default high thinking level, according to Google, a figure also reported by Ars Technica and TechCrunch.
• Code: the Terminal-Bench 4.1 score rises from 18.2% for Gemini 3.8 Flash to 57%, more than triple, according to Google DeepMind.
• Reliability: fabricated claims drop by more than a factor of five, according to Google's own analysis.
• Partner tests: on graph, SQL and data-heavy enterprise workflows, partners measured runs 2 to 2.8 times faster and 12% to 23% fewer tokens than with 3.8 Flash, according to Google.
• Context: native 1 million-token input and up to 1 million tokens of output, according to Google DeepMind.

On the independent Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high setting against 50 for Gemini 4 Argon, and 54 at the new xhigh setting, according to Google. The model has four thinking levels, namely low, medium, high (the default) and xhigh. They let developers trade quality against latency and cost. These are figures published by Google, and no third-party reproduction was cited in the sources we read.

Google also lists results on specialised tests. According to Google DeepMind, Hydrogen scores 58.8% on HLE-Diamond, a hard-reasoning benchmark where it leads its class. It ranks first on the Vals Tax Agent Bench at 52.3%, ties for first on the Vals Legal Research Bench at 55.8%, and ranks fifth on arena.ai's live Agent Arena. It also scores 53.4% on the Multimodal Knowledge Agent benchmark and 71.6% on Agents' Last Exam v1.1.

How much does Gemini 4 Hydrogen cost?

Google prices Hydrogen at a flat $1.50 per million input tokens and $7.50 per million output tokens across its entire 1 million-token context window, according to Google DeepMind. Ars Technica adds that this is unchanged from Gemini 3.8 Flash. The same article reports that Carbon and Argon currently cost $2 and $10 per million tokens, and will rise to $4 and $20 from next month. If that schedule holds, the gap between the frontier models and Hydrogen will widen sharply.

What changes in Google Search?

Google describes the move as the biggest model upgrade to Search yet, according to both Google's Keyword blog and TechCrunch. Hydrogen is rolling out to all Search users globally, and the changes concern AI Mode and AI Overviews. In its post, Google lists four improvements:

• Better understanding of intent: questions with several parts, hard-to-understand typos or very specific requests are handled more precisely.
• Improved query fan-out: this technique splits one prompt into several simultaneous sub-queries, and Google says it now links to a wider range of sites.
• More visual answers: more images, structured responses, comparison tables and custom tools.
• Richer language understanding: more nuanced answers in all languages supported by AI Mode.

Ars Technica is cautious about the linking claim and writes that it has heard similar promises from Google before. Whether the new fan-out sends more traffic to independent sites remains unproven, and none of the sources we read provides figures on clicks or referrals.

What is Orchestrator in Google Antigravity?

Alongside the model, Google introduced Orchestrator in Antigravity. Instead of using one model from start to finish, Orchestrator switches dynamically between models such as Carbon and Hydrogen within the same session. On Terminal-Bench 4.0, Google says it reached near-Carbon task success while using on average 24% less quota and finishing 19% faster. For parallel agents, Carbon can coordinate while Hydrogen powers the worker and verifier agents that build and test changes in isolated workspaces.

What is still unclear?

Access is the main point on which the sources diverge. Google DeepMind says Hydrogen reaches all tiers of AI Studio and the Gemini app. Ars Technica reports instead that the model is live for AI Pro and Ultra subscribers, while AI Plus and free users remain on the older 3.5 Flash-Lite model. We treat the exact availability per plan as unconfirmed. All performance numbers above come from Google and from Google's reading of third-party benchmarks.

What it changes for a company

For a company that already builds on Gemini 3.8 Flash, the switch looks straightforward on paper: the price per million tokens is unchanged according to Ars Technica, while speed and code scores rise according to Google. The sensible step is to rerun your own test set on the low, medium and high levels before migrating, since the published figures are vendor-reported. Teams that route work between several models can also test the split Google proposes, with a frontier model for design decisions and Hydrogen for execution and verification. Finally, marketing and content teams should watch their Search referral data in the coming weeks, because answers with more tables and links change what users click. For the agent architectures behind such routing, see our guide to the agentic system.