Gemini 4 Hydrogen: Google's Fast Everyday AI Model, and the Biggest Search Upgrade Yet
On 12 November 2026, Google introduced Gemini 4 Hydrogen, the third model in its Gemini 4 family โ and the one most people will actually meet. While Gemini 4 Carbon and Gemini 4 Argon are Google's frontier models for deep, long-horizon reasoning, Hydrogen is built for rapid execution: everyday coding, agentic workflows, and โ most consequentially โ Google Search, where it is now rolling out to every user worldwide.
It is also strikingly cheap, surprisingly strong on benchmarks for a "fast" model, and โ like every model launch โ best understood with its numbers attributed to their source.
Where Hydrogen sits in the Gemini 4 family
Google's framing is a division of labour. Carbon and Argon handle the hardest problems; Hydrogen does the fast work โ writing code, running tests, verifying changes, carrying out multi-step tasks. It is the model Google describes as a major standalone upgrade for teams migrating from Gemini 3.8 Flash, the previous fast model, and it takes over the everyday role across the Gemini app, the Gemini API, Google AI Studio and Search.
The surprise in Google's own announcement is that the fast model is not obviously weaker. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high thinking level, against 50 for the larger Argon โ as reported by Google โ and reaches 54 at a new xhigh setting, the fourth of Hydrogen's thinking levels (low, medium, high and xhigh). Google also says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at that default setting, more than triples its Terminal-Bench 4.1 score (57% against 18.2%), and cuts fabricated claims by more than five times.
Those figures are Google's own, measured on Google's own harnesses, and they have not been independently replicated. The Artificial Analysis index is an independent benchmark, but the 51 and 54 scores are as reported by Google. That caveat applies to every benchmark number in this article.
Google also cites leading positions on several agent benchmarks โ fifth place on Agent Arena's live ranking, 58.8% on HLE-Diamond, first place on the Vals Tax Agent Bench at 52.3% โ all, again, as reported by Google.
What it costs
Hydrogen is priced flat across its entire 1-million-token context window: $1.50 per million input tokens and $7.50 per million output tokens. There is no price step for longer context, and the window pairs 1 million input tokens with up to 1 million output tokens.
For developers, the price matters because Hydrogen is meant to be used in volume โ as the execution engine in multi-model workflows, where Carbon plans and Hydrogen does.
Orchestrator: two models in one session
The most interesting product idea in the announcement is not Hydrogen alone but what it enables in pairs. Google is introducing Orchestrator in Antigravity, its agentic development platform, which dynamically switches between models โ Carbon and Hydrogen โ inside a single session. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon success rates while using 24% less quota and finishing 19% faster. Carbon can also act as coordinator while Hydrogen powers worker and verifier agents in parallel, building and testing changes in isolated workspaces.
Orchestrator is available to paid Google AI Pro and Ultra subscribers in Antigravity โ that is the availability Google's announcement states โ and its benchmark is, again, Google's own.
What changes in Search
For most readers, the Search rollout is the news that matters. Robby Stein, Google's vice president of product for Search, calls Hydrogen "the biggest model upgrade to Search yet", and says it is now available globally in Google Search for all users. The upgrade also brings more highly nuanced responses across all supported AI Mode languages.
What Google says changes: Search understands intent better โ multi-part questions, typos, highly specific requests โ gives more visual responses (images, tables, comparison tools), and fans queries out across more subtopics, connecting users to a broader range of links than before. Hydrogen also works across all supported AI Mode languages.
Two things Google has not provided: any before-and-after measurement of Search quality โ no error rates, no human-rater scores, no third-party audit of AI Mode or AI Overviews. The "biggest upgrade" claim is Google's characterization, not a measured result, and readers will have to judge the improvement from using it.
Who actually gets Hydrogen
Rollout is broad but tiered, and the details matter. Hydrogen is available in all tiers of Google AI Studio and the Gemini API, in Gemini for business, in the Gemini app, in Google Search for all users, and โ for paid Google AI Pro and Ultra subscribers โ in Antigravity and in Gemini in Google Sheets.
In the Gemini app, Hydrogen sits in the model drop-down for everyone, and Google says it is bringing Gemini 4 to everyone โ across the app, Google Search, the API and AI Studio.
The safety numbers โ and what they actually mean
Because Hydrogen is built for agentic work โ a model that takes actions, calls tools and spends your tokens โ Google puts unusual weight on its safety figures. The announcement claims Hydrogen reduces overreach in agentic tasks by up to 81% and under-service by up to 79%, cuts misaligned actions and task shortcutting across tool-use and coding evaluations by 12 times, and achieves a 94% honesty rate on Google's internal agent honesty evaluation. It preserves 99.9% chain-of-thought legibility with zero deceptive reasoning edits across more than 200,000 audited turns, is more than 2.3 times more resilient to indirect prompt injection on the Gray Swan benchmark, and keeps false refusals on HealthBench Pro under 0.4%.
Those are strong numbers, and they are all internal Google evaluations. No independent safety audit accompanies the launch, and the benchmarks that sound most absolute โ "zero deceptive reasoning edits" โ describe Google's own audit of its own model. Treat them as a statement of what Google tested, not as independent proof.
The bottom line
Gemini 4 Hydrogen is the Gemini 4 model designed to disappear into everyday use: fast, cheap, in Search for everyone, and built to do the routine work while Carbon and Argon think about the hard problems. Its headline numbers are impressive โ a fast model that outscores its bigger sibling on one independent index, at a flat $1.50 per million input tokens โ but every one of them comes from Google, and the Search improvement it promises has no public measurement yet. The honest summary is narrower than the launch language: a genuinely useful new default for everyday AI work, whose real quality will be settled by independent testing and by what users notice in Search, not by the announcement.
CurrentBeam uses AI-assisted research and drafting. Our editorial guidelines explain our sourcing, review process, and how to request corrections. Editorial guidelines
Help improve this article
Found an error or missing context? Send the article link and supporting source to our editorial team.
Email a correction