Advertise hereReach developers building with AI coding agents. →
Cursor Coding

One day after Fable 5.2 took #1, Gemini 4 Carbon edged past it on CursorBench at 60.4%

Cursor added Google's Gemini 4 Carbon to its model list on November 10 and says that at high effort it is now the top model on CursorBench, Cursor's own coding benchmark, with 60.4%. That nudges past Claude Fable 5.2, which Cursor had put on top a day earlier with 60% at max effort (56.9% at default). Google's Antigravity team also cites the CursorBench 4.0 ranking for Carbon.

Why it matters

The top of the coding leaderboard changed hands in about 24 hours, and the gap is 0.4 points. For real work, the better test is your own repo, and the price per finished task, not the headline score.

Who it's for

Cursor users who pick models per task, and teams deciding between Gemini and Claude for agent runs.

How to try it

  1. Update Cursor and open the model picker in the agent panel.
  2. Choose Gemini 4 Carbon and set effort to high for a hard task.
  3. Run the same task with Claude Fable 5.2 and compare the diff, time and cost.
  4. Keep a cheaper model as default and switch to Carbon only for tough jobs.

Plan & availability: In Cursor's model picker from November 10, 2026; usage follows your Cursor plan and model pricing. The 60.4% score is Cursor's own benchmark result at high effort.

Share on X

Source: @cursor_ai on X ↗

Written by AgentStackList from the original post. Checked Nov 10, 2026. Features, plans and availability can change, so confirm on the source before you rely on it.

cursorgemini-4-carboncursorbenchbenchmarks

More Cursor tricks

All Cursor tricks →

Agent Stack Weekly · free

Get the week's best AI tricks

Limit and pricing changes, new MCP servers, and setup fixes for Claude Code, Codex, Cursor, Copilot and Antigravity.

No spam. Unsubscribe anytime. Read past issues