Google's new fast model jumped almost 15 points on CursorBench: Gemini 4 Hydrogen is now in Cursor
Cursor added Gemini 4 Hydrogen to its model list on November 12, 2026, the same day Google launched it. Cursor says that at its default effort Hydrogen scores 54.3% on CursorBench, Cursor's own coding benchmark, up from 39.6% for the previous fast Gemini, 3.8 Flash. For comparison, Cursor put the much heavier Gemini 4 Carbon at 60.4% on high effort earlier this week.
Why it matters
A fast, cheaper model that lands within about six points of the current leaders is a good candidate for your everyday default, saving the big models for the hard jobs.
Who it's for
Cursor users who want a quick, lower-cost model for routine edits, tests and small features.
How to try it
- Update Cursor and open the model picker in the agent panel.
- Select Gemini 4 Hydrogen and give it a routine task, like adding tests to one file.
- Run the same task on your usual model and compare the diff, time and cost.
- If it holds up, make Hydrogen your default and switch up only for tough, multi-file work.
Plan & availability: In Cursor's model picker from November 12, 2026; usage follows your Cursor plan and model pricing (Google lists Hydrogen at $1.50/$7.50 per million tokens in its own API). The 54.3% is Cursor's own benchmark result at default effort.
Source: @cursor_ai on X
Written by AgentStackList from the original post. Checked Nov 12, 2026. Features, plans and availability can change, so confirm on the source before you rely on it.
cursorgemini-4-hydrogencursorbenchbenchmarks
More Cursor tricks
- Today is the day OpenAI picked to pull GPT out of Cursor: what actually breaks, and 3 ways to keep using it · Nov 12
- One day after Fable 5.2 took #1, Gemini 4 Carbon edged past it on CursorBench at 60.4% · Nov 10
- Your coding agents can now talk to your teammate's agents: Cursor Projects goes multiplayer · Nov 10
- Claude Fable 5.2 lands in Cursor and takes the top spot on CursorBench at 60% · Nov 9