Google equips Gemini with a faster and more direct Hydrogen model
Google replaces the Flash branding with Gemini 4 Hydrogen, presented as faster and more reliable. The model powers the Gemini app, Search, and Antigravity, with announced gains on agentic tasks.
Google unveiled Gemini 4 Hydrogen, a new AI model presented on November 12, 2026, as a clear step up from Gemini 3.8 Flash. The company highlights gains in speed, reasoning, and reliability, with a direct impact in the Gemini app, Google Search, and Antigravity. For daily uses as well as agentic work, the change matters because it affects both response quality and execution cost.
A model designed to go faster without losing track
Hydrogen is described by Google as a “major standalone upgrade”, with “significant leaps in both speed and intelligence.” The model natively accepts 1 million input tokens and up to 1 million output tokens, which greatly expands the range of documents, conversations, and codebases it can process in one go.
Regarding performance, Google announces tasks executed 2.5 times faster at its default high setting, more than three times better on Terminal-Bench 4.1, and a reduction in “fabricated claims” by more than 5 times. The company also says Hydrogen adapts better when a tool call fails or a search returns nothing. In other words: fewer answers going off the rails. That’s never useless.
In Gemini, Search, and Antigravity, orchestration becomes the focus
In the Gemini app, Hydrogen is supposed to better follow the user's tone, format, and instructions. Google promises more readable tables, cleaner summaries, and drafts of emails, study guides, or documents “ready to use on the first try.” The model would also more often be able to choose the right tool or connected skill.
The model selector places Hydrogen in the “Everyday help” category, between 3.5 Flash-Lite for quick answers and 4 Carbon for the most complex tasks. Free Gemini users access it via automatic selection when the system judges it the right model for the query.
On Search, Hydrogen becomes the default model for AI Mode and also serves for the toughest questions in AI Overviews. Google says the system better understands naturally formulated queries, including those with multiple parts, typos, or very precise requests. Answers should also become more visual, with more relevant links.
The real message from Google: reducing agent costs
The most interesting part may be in Antigravity, Google's workspace for AI agents. Hydrogen receives an “Extra” reasoning level alongside Low, Medium, and High. Most importantly, Google introduces Orchestrator, a mechanism that distributes tasks among several models instead of having one do everything.
In this scheme, Carbon can coordinate, while Hydrogen handles work and verification agents in isolated spaces. The goal is clear: advance several steps in parallel without exploding the token bill, those text units used to count model input and output.
For developers, Google announces a price of $1.50 per million input tokens and $7.50 per million output tokens. Hydrogen is reserved in Antigravity for paying Google AI Pro and Ultra subscribers. Google also says its partners have seen the model run 2 to 2.8 times faster while reducing token consumption by 12% to 23% compared to 3.8 Flash. That’s where the cost becomes concrete: less cost, less waiting, and potentially less tinkering on multi-agent workflows.
Key figures summarizing the shift
- 1 million input tokens, natively.
- 2.5 times faster, according to Google, at default high setting.
- 96.7% honesty rate on internal evaluation cited by Google.
- $1.50 per million input tokens for developers.
- $7.50 per million output tokens for developers.