Cursor: Gemini 4 Carbon Tops CursorBench at 60.4 Percent
Gemini 4 Carbon is now available in Cursor! At high effort, it's the top model on CursorBench at 60.4%
Context
A top score on Cursor's own coding bench, plus immediate access inside Cursor, is an early real world test of Google's new model.
Cursor posted that Gemini 4 Carbon is now available in Cursor, and that at high effort it is the top model on CursorBench at 60.4%. Google opened Gemini 4 Carbon earlier today, a launch also covered in this deck. The chart attached to Cursor's post compares Carbon with other models on CursorBench 4.0 by score and cost.
Deep dive
Cursor posted that Gemini 4 Carbon is now available in Cursor. The same post said that at high effort, Carbon is the top model on CursorBench at 60.4%.
The attached chart is labeled CursorBench 4.0 score and plots models by score and dollar cost. Carbon at high effort is marked at 60.4% with a cost of $9.22. Other models visible on the chart include Fable 5.2, Gemini 4 Argon, Opus 5.5, Muse Wonder, Grok 4.7, Sonnet 5.5, Haiku 5.5, and GPT 5.6 Sol. The chart is Cursor's own presentation of the bench, not an independent audit.
Google posted earlier today that Gemini 4 Carbon is now available and called it the company's strongest model. That launch is covered separately in this deck. Cursor's post is a product availability note plus a benchmark claim from CursorBench, and it does not give pricing for Carbon inside Cursor, a full leaderboard table, or a definition of high effort beyond the label on the chart.
What to expect next
Watch for whether Carbon holds that CursorBench lead as more teams try it on live repos, and for Cursor to publish a fuller leaderboard or a definition of high effort. Independent benches and developer reports will show whether the 60.4% result holds outside Cursor's chart. Pricing and rate limits inside Cursor, which the launch post does not include, will also decide how widely the model gets used.