CursorBench 4.1: leaderboard

Cursor coding-agent benchmark of ambiguous, multi-file tasks from real Cursor sessions, scored by task success percentage with average cost, tokens and steps per task reported separately. Each published reasoning-effort run remains separate. Released on 2026-11-18, the 4.1 task set modifies instruction-following problems and increases the agent time limit relative to CursorBench 4.0 and is not comparable with earlier CursorBench scores.

Metric: Score (%). Source: cursor.com. Saturation forecast: Around July 2028. 12 models tracked. Updated 19 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 4 NeonxHigh65.4#3
2Gemini 4 CarbonHigh61.5#5
3Gemini 4 HydrogenxHigh61.4#16
4Claude Fable 5.2Max61.3#1
5Claude Opus 5.5Max58.1#9
6Gemini 4 ArgonHigh56.5#18
7Claude Sonnet 5.5Max54.4#24
8Grok 4.7xHigh48.6#51
9Claude Haiku 5.5Max48.4#75
10Grok 4.6xHigh44.3#50
11Muse Spark 1.3Max41.9#48
12GPT-5.6 SolMax40.9#36

No result here: #4 GPT-6.2 Astra, #10 GPT-6 Astra.

Interactive version: theaggregate.ai/benchmark?slug=cursorbench-4-1 · How It Works · Data refreshed daily, snapshot 2026-11-20.