opus-4.6 vs claude-opus-4-5
Both models ran the same corpus under the same verifier — cursor and opencode respectively. The harness differs where the names differ, and that is part of what separates them.
opus-4.6 leads by 25.7 points, and the intervals do not overlap.
Where they differ
| Weakness | opus-4.6 | claude-opus-4-5 | Difference |
|---|---|---|---|
| CWE-125Out-of-bounds Read | 65.3% | 41.9% | +23.4 |
| CWE-787Out-of-bounds Write | 76.5% | 41.2% | +35.3 |
| CWE-416Use After Free | 46.2% | 23.1% | +23.1 |
| CWE-908Use of Uninitialized Resource | 33.3% | 25.0% | +8.3 |
| CWE-415Double Free | 66.7% | 33.3% | +33.4 |
| CWE-1284Improper Validation of Specified Quantity in Input | 50.0% | 50.0% | 0.0 |
| CWE-843Access of Resource Using Incompatible Type ('Type Confusion') | 50.0% | 0.0% | +50.0 |
| CWE-476NULL Pointer Dereference | 50.0% | 0.0% | +50.0 |
| CWE-475 | 100.0% | 0.0% | +100.0 |
| CWE-562 | 100.0% | 0.0% | +100.0 |
Per-class samples are small; a difference here is a hint about where to look, not a result. The corpus-level intervals above are the honest comparison.
Corpus cve-bench-136 · run February 2026 · How we test
See the benchmark →