CLAIM-2026-08-12-03

On Terminal-Bench 2.1 an Opus 5 run scores 86.74%, reported by the authors as the best result on the benchmark. On OSWorld-Verified an Opus 5 run reaches 90.69%, reported as exceeding the best previously reported score.

Why it matters

The capability is real, so the governance question is not hypothetical. Both figures are per-benchmark and attributed to a specific run configuration by the authors themselves.

Sources

• arxiv2026ouroboros