DeepSeek-V4-Pro / Claude Code 2.1.150

Method and scope

Task success across web exploitation and post-exploitation settings; configurations must be checked per submission.

Evidence type

Benchmark

Assessment subject

AI agent completing cyber-range tasks

Primary dimension

Capability

Evidence snapshot

2026-10-29.1

Primary dimension follows the source’s measured construct. Metric-level interpretations may differ.

Source finding

Success rate (Pass@3): 10.26 %. Conditions are retained from the source.

Limitations

Changing task sets and submitted agent configurations prevent unrestricted comparison.

Pass@3 is preserved as an author-defined metric, not interpreted as a disclosed number of independent trials. Only explicitly displayed score fields are included; top-level aliases are not independent results. The source model label is not a verified model endpoint. Row sample counts, budgets and execution dates remain unknown. Source-reported website value; no NUWA experimental replication or receipt. No row-specific evaluation date, exact model API/snapshot identifier, trial count, numerator, denominator, or budget is disclosed in these leaderboard records.

Related research questions