All Benchmarks/Agents

CursorBench

Often citedHigher is better

Tasks collected from real sessions of the Cursor coding tool: ambiguous, multi-file work such as understanding code, finding bugs, editing, refactoring, and code review. Cursor runs it on its own product's tasks, and scores from different versions are not comparable. Scores run from 0 to 100%. Higher is better.

Top score57.8%Claude Opus 5.5
Models tested14
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    Anthropic
    Claude Opus 5.5
    2026.09.22 · Max
    57.8%
  2. 2
    Anthropic
    Claude Sonnet 5.5
    2026.09.28 · Max
    55.5%
  3. 3
    Anthropic
    Claude Fable 5.1
    2026.09.01 · Max
    51.8%
  4. 4
    Anthropic
    Claude Opus 5
    2026.07.24 · Max
    46.6%
  5. 5
    xAI
    Grok 4.7
    2026.09.21 · Extra High
    46.3%
  6. 6
    Z.ai
    GLM 5.3
    2026.08.18 · Max
    42.6%
  7. 7
    OpenAI
    GPT-5.6 Sol
    2026.07.09 · Max
    41.7%
  8. 8
    Meta
    Muse Spark 1.3
    2026.09.02 · Max
    41.6%
  9. 9
    xAI
    Grok 4.6
    2026.08.12 · Extra High
    41.4%
  10. 10
    OpenAI
    GPT-5.6 Terra
    2026.07.09 · Max
    41.3%

The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.