NUWA Evidence record
DeepSeek V4 Flash
EvalAwareBench · Evidence record
Source snapshot, chart and review scope →
Original source charts and materials
Recorded measurements
- Recognition
- 6.4%
Per-result sample size not disclosed
Harness not specified · Budget not specified
Assessment provenance
Evaluating Kimi K3 for Misalignment Risks → · External report
Actual evaluation: Neo Research · Model record →
Method and scope
Varies evaluation cues while tracking verbalized evaluation awareness in controlled tasks.
- Evidence type
- Benchmark
- Assessment subject
- The model or agent in the stated task configuration.
- Primary dimension
- Behavior
- Evidence snapshot
- 2026-10-29.1
The main dimension follows the stated measurement; contextual risk relevance does not establish harmful deployment capability.
Source finding
Recognition: 6.4 %. Conditions are retained from the source.
Limitations
Unexpressed awareness is not measured; recognition of familiar benchmarks can differ from general cue response.
Flash exceeded token limit before completion on 52% of tasks. Safety-task completion means compliance under benchmark conditions, not general safety.
Related research questions
Deception / Evaluation gaming →
The protocol explicitly measures this question or an enabling sub-capability; conclusions retain its stated conditions.
References
- [1] EvalAwareBench
- [2] Evaluating Kimi K3 for Misalignment Risks. 2026-09-11. September 2026 ; §2.4.2 Table 6 p.30