CursorBench 3.2 — GPT‑5.6 expanded

Score by average cost per task. Every published Sol, Terra, and Luna reasoning level is included.

Source ↗
Higher and farther left is better
CursorBench 3.2 score by average cost per task Scatter plot comparing benchmark score and cost, including all GPT-5.6 Sol, Terra, and Luna reasoning efforts.

Published CursorBench values captured July 13, 2026. Small score differences may not be statistically meaningful. Lines connect reasoning levels within a model family.

Configured subagents: Sol Extra High · Terra Extra High · Luna High