The speed-focused model scored 51 on the Artificial Analysis Intelligence Index, one point above Google's frontier Argon model at default settings
Google DeepMind has launched Gemini 4 Hydrogen, a model that finishes tasks 2.5 times faster than Gemini 3.8 Flash. It also outscored Gemini 4 Argon, Google’s own frontier model, on the Artificial Analysis Intelligence Index.
Hydrogen was announced on November 12, 2026, and is rolling out across all Gemini tiers immediately. It is live in the Gemini app, the Gemini API, Google AI Studio, Vertex AI and Google Search.
The numbers behind Hydrogen #
The headline speed figure comes from Hydrogen’s default high thinking setting, where Google says the gains are most pronounced. Compared with Gemini 3.8 Flash, the new model completes tasks 2.5 times faster.
On the Artificial Analysis Intelligence Index, Hydrogen scored 51 at default settings. Argon, the frontier model, scored 50 under the same conditions.
Push Hydrogen to its highest capacity and the score climbs to 54.
The coding results show the biggest jump. On Terminal-Bench 4.1, Hydrogen posted a score of 57%, up from 18.2% for its prior version.
Terminal-Bench measures how well a model handles tasks inside a command-line environment, the text-based workspace developers use to run code and manage systems.
Where it fits in the Gemini lineup #
Google is positioning Hydrogen as the fast, practical member of the Gemini 4 family. Its intended jobs are rapid execution, algorithm coding and general-purpose work.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The heavier lifting is meant to stay with Gemini 4 Carbon and Argon, the more capable frontier models. Hydrogen is designed for multi-modal orchestration alongside them, so the three can split a workload rather than compete for it.
Pricing is a flat rate of $1.50 per million input tokens and $7.50 per million output tokens. Tokens are the small chunks of text a model reads and writes, so input is what you send in and output is what comes back.
The model carries a context window of 1 million tokens.
Google says Hydrogen is aimed at developers and high-volume users.
A quiet testing run before launch #
Hydrogen did not arrive out of nowhere. Community testing ran from late October 2026 under the name gemini-3.8-flash, which meant some users were trying the model before it had its real name.
Internal staff testing was reported on November 11, 2026, one day before the public announcement.
What this means for developers and the AI race #
For developers, the most practical story is the combination of speed, coding gains and a predictable flat price. A model that finishes tasks 2.5 times faster can change how responsive an app feels, especially in products where users sit and wait for an answer. The Terminal-Bench jump matters most for teams building coding assistants and automated agents. A model that succeeds on 57% of command-line tasks instead of 18.2% is far more usable for work that involves actually running things, not just writing snippets.
The Intelligence Index result raises a more awkward question for Google’s own lineup. If the fast tier scores 51 and the frontier tier scores 50 at default settings, buyers may wonder when the bigger model is worth reaching for.
Google’s answer appears to be orchestration. Hydrogen is built to work alongside Carbon and Argon, so the pitch is less about picking one model and more about routing each task to the right tool.
Pricing will be worth watching as usage grows. A flat rate of $1.50 per million input tokens and $7.50 per million output tokens is easy to budget, but the five-to-one gap between output and input means cost depends heavily on how chatty an application is.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.