NUWA Evidence record
Kimi K3
Cybench · Evidence record
Source snapshot, chart and review scope →
Original source charts and materials
Recorded measurements
- Pass@1
- 97.4%
38/39
Harness not specified · 3 submission attempts/challenge; unlimited wall-clock
Assessment provenance
Evaluating Kimi K3 for Misalignment Risks → · External report
Actual evaluation: Neo Research · Model record →
Method and scope
Task completion on 40 professional CTF problems.
- Evidence type
- Benchmark
- Assessment subject
- AI agent solving CTF 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
Pass@1: 97.4 %. Conditions are retained from the source.
Limitations
Competition tasks and supplied guidance differ from operational attacks.
Text calls tasks unguided while Figure 12 calls this a hard subset. Not independent replication; contaminated/saturated tasks may inflate performance.
Related research questions
Cybersecurity / Cyber task execution →
The source directly measures the stated construct under its protocol.
References
- [1] Andy K. Zhang, Neil Perry, et al.. Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models. 2025-04-12. v4 [arXiv]; Author affiliations on the original paper title page
- [2] Evaluating Kimi K3 for Misalignment Risks. 2026-09-11. September 2026 ; §2.1.1 pp.19–20 Figure 12