NUWA Evidence record
DeepSeek v4 Pro
EVMbench · Evidence record
Source snapshot, chart and review scope →
Original source charts and materials
Recorded measurements
- Exploit
- 32.7%
Per-result sample size not disclosed
Harness not specified · Budget not specified
Assessment provenance
Evaluating DeepSeek v4 Pro for Frontier Risks → · External report
Actual evaluation: Neo Research · Model record →
Method and scope
Separates detection, patching and exploitation modes with their stated task sets.
- Evidence type
- Benchmark
- Assessment subject
- The model or agent in the stated task configuration.
- Primary dimension
- Capability
- Evidence snapshot
- 2026-10-29.1
The main dimension follows the stated measurement; contextual risk relevance does not establish harmful deployment capability.
Source finding
Exploit: 32.7 percent. Conditions are retained from the source.
Limitations
Modes and task subsets cannot be combined into an undifferentiated cyber-risk score.
Numeric labels read directly from rendered chart.
Related research questions
Cybersecurity / Cyber task execution →
The protocol explicitly measures this question or an enabling sub-capability; conclusions retain its stated conditions.
References
- [1] EVMbench
- [2] Evaluating DeepSeek v4 Pro for Frontier Risks. 2026-05-29. v1.1 · 29 May 2026 ; PDF p. 9; EVMbench; Exploit