Google has launched Gemini 4 Hydrogen, a fast execution model that immediately becomes the default in AI Mode and the engine behind the hardest questions in AI Overviews. Announced on November 12, 2026, the model is positioned as a cheaper, quicker sibling to Google's frontier Gemini 4 systems, with a one-million-token context window and flat per-token pricing. It is rolling out globally across Search, the Gemini app, the API and Google's developer tools.
Key Takeaways
- New default: Hydrogen takes over as the default model in AI Mode and handles complex queries in AI Overviews.
- Speed claim: Google says it completes tasks 2.5 times faster than Gemini 3.8 Flash at the default high thinking level.
- Pricing: $1.50 per million input tokens and $7.50 per million output tokens, flat across the full context window.
- Context: native 1 million tokens of input and up to 1 million tokens of output.
- Availability: free Gemini users get it through automatic model selection; paid AI Pro and Ultra subscribers get direct access.
What Gemini 4 Hydrogen Actually Is
Hydrogen is not Google's most capable model, and the company does not present it as one. Gemini 4 Carbon and Gemini 4 Argon remain the systems for long-horizon reasoning and difficult architectural problems. Hydrogen is built for execution: writing code, running tests, checking changes, calling tools and finishing multi-step tasks without consuming frontier-level compute.
That distinction is the point. Google argues that most everyday work does not need the largest model available, and that a faster, cheaper system with disciplined instruction-following produces better results per dollar. Hydrogen is designed to run alone or as the execution layer in workflows coordinated by a frontier model.
The naming is less clear than the strategy. The "Flash" label that marked Google's fast tier is gone, replaced by element names that reveal nothing about capability or speed.
The Search Upgrade Is the Headline
Search is where most people will encounter the model, and Google is treating the swap as a significant change. Robby Stein, vice president of product at Google Search, described it as "the biggest model upgrade to Search yet" and confirmed Hydrogen is now the default in AI Mode and the model used for the hardest AI Overviews queries.
Three changes matter in practice. First, intent parsing: Google says AI Mode can now handle questions written in a natural flow of thought, including multi-part requests, severe typos and highly specific constraints, and address each part separately rather than collapsing them into one answer.
Second, retrieval. Hydrogen upgrades the query fan-out technique that breaks a single prompt into multiple subqueries. Google says this connects users to a broader range of links, including sites they might previously have missed — a detail that matters to publishers who have watched AI summaries absorb clicks.
Third, presentation. Responses in AI Mode will lean harder on visuals: comparison tables, graphs, maps, images and interactive tools generated on the fly. Google's own example is an animated driving simulation explaining how phones determine a car's lane without cell service. The company also says language understanding improved across every AI Mode language.
Benchmarks, Context and Price
The published numbers are company-reported and compare Hydrogen with Gemini 3.8 Flash, its predecessor in Google's fast tier.
| Metric | Google's claim |
|---|---|
| Task speed | 2.5× faster at the default high thinking level |
| Terminal-Bench 4.1 | 57%, up from 18.2% |
| Context window | 1M input tokens; up to 1M output tokens |
| API pricing | $1.50 / 1M input, $7.50 / 1M output |
| Fabricated claims | Reduced by more than 5× |
| Agent honesty | 96.7% on internal evaluation |
| Thinking levels | Low, medium, high (default), xhigh |
Google also claims Hydrogen reaches 51 on the Artificial Analysis Intelligence Index at its default setting, slightly ahead of Gemini 4 Argon at 50, and 54 at the new xhigh tier. It reports 58.8% on HLE-Diamond for deep domain reasoning, 53.4% on an agentic knowledge-work benchmark involving visual files, and 71.6% on a professional-workflow evaluation.
The pricing may matter more than any single score. Charging one rate across the full million-token window removes the long-context penalty that makes document and codebase work expensive, and pressures rivals that tier prices by context length.
Orchestrator and the Multi-Model Bet
Hydrogen arrives alongside Orchestrator, a routing system in Google Antigravity that switches between models inside a single session instead of committing to one from start to finish. Routine steps can be handed to Hydrogen while Carbon handles planning or difficult judgment calls.
Google says Orchestrator reached near-Carbon task success rates while using 24% less quota and finishing 19% faster. In parallel-agent setups, Carbon can act as coordinator while Hydrogen powers worker and verifier agents building and testing changes in isolated workspaces.
That is the real strategic shift: Google is no longer selling one model as the answer to everything, but a portfolio where the cheap model does most of the work and the expensive one supervises. Whether developers trust automatic routing over explicit model choice will decide how much of it sticks.
Who Gets It, and Where
- Search: default in AI Mode and used for complex AI Overviews queries, globally.
- Gemini app: free users get it through automatic model selection; AI Pro and Ultra subscribers can pick it directly.
- Developers: all tiers of AI Studio, the Gemini API, Gemini for business, and Antigravity for paid subscribers.
- Workspace: Gemini in Sheets for eligible business subscriptions, where Google says financial models generate over 30% faster.
What to Watch Next
Search quality in the wild. The fan-out change has the clearest consequences for publishers: more links cited, or fewer clicks per answer, depending on how the interface evolves.
Independent benchmarks. Most of the published figures come from Google. Third-party evaluations of Terminal-Bench, agentic reliability and long-context accuracy will be more informative than launch-day comparisons.
Routing behaviour. If Orchestrator quietly sends most requests to Hydrogen, the practical gap between the fast and frontier tiers may narrow faster than the naming suggests.
The next Carbon release. Hydrogen sets a new baseline for speed and price. The frontier model that follows will have to justify its premium against a cheaper system that now handles the default workload.
Follow Us on Facebook
