← Claude Sonnet 5 · API · High effort (default)
Evidenceacceptedthird_party_leaderboard
CursorBench
Cursor CursorBench (reported via launch coverage citing Cursor independent measurement)
ev_2026_07_sonnet5_cursorbench
Result
61.2%
Confidence B
- Tested
- 2026-06-30
- Runs
- 1
Configuration
How it was measured
- Exact model
- claude-sonnet-5
- Reasoning
- high
- Agent harness
- Cursor_production_CursorBench
- Tool access
- IDE agent tools (Cursor)
- Prompting
- Cursor production agent harness
Independent IDE-harness signal (not SWE-bench). Stretch STR / existing_codebase.
Caveats
Known limitations
- Primary figure is widely cited at launch; prefer Cursor's own published post when verifying freeze.
- Different task distribution than SWE-bench Verified.
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.85
quality0.80
recency1.00
health0.75
comparability0.70
reliability0.70
Benchmark
CursorBench
Independent coding preference / success evaluation measured by Cursor in their production IDE agent harness (not SWE-bench).
Benchmark source →