Composer 2.5: benchmark results
Provider: Cursor. Released 2026-05-18. Access: API.
Aggregate Rating 1746 ± 20, rank #122 of 1656 rated models, from 16 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Vals AI SWE-bench Verified | 79.6 | Resolved (%) | 71.3 |
| Agent Security League - Functional Correctness | 75.4 | Functional Correctness (%) | 60.9 |
| InfraBench - Attempt Pass@0.5 | 75 | Attempts scoring at least 0.5 (%) | 52.9 |
| Vals AI Vibe Code Bench | 49.61 | Accuracy (%) | 50.9 |
| InfraBench - Mean Effective Score | 72.6 | Difficulty-weighted mean effective score (0-100; mean of thr | 47.1 |
| Vals AI Terminal-Bench 2.1 | 58.43 | Accuracy (%) | 45.9 |
| InfraBench - Attempt Pass@1 | 58.3 | Attempts with a perfect score (%) | 41.2 |
| InfraBench - Best-of-3 Perfect Tasks | 66.7 | Tasks with a perfect best-of-three attempt (%; of 12) | 35.3 |
| FrontierSWE | 34 | Dominance (%) | 34.4 |
| Agent Security League - Security Correctness | 14 | Security Correctness (%) | 30.4 |
| KernelBench Mega | 2.48 | Speedup (x) | 10.7 |
| CursorBench 4.0 | 27.7 | Score (%) | 9.5 |
Interactive version: theaggregate.ai/model?slug=composer-2-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.