NUWA Evidence record
DeepSeek V4 Flash
CharXiv · Evidence record
Source snapshot, chart and review scope →
Original source charts and materials
Recorded measurements
- Attempt rate
- 48.2%
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
Tests understanding and reasoning over charts. Neo’s withheld-image adaptation measures attempts without the required input.
- 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
Attempt rate: 48.2 %. Conditions are retained from the source.
Limitations
The original visual capability benchmark and the missing-image behavioral test measure different constructs.
Attempt includes correct prior-knowledge guesses and incorrect answers; not original CharXiv capability accuracy. Figure depicts confidence intervals and correct/incorrect stacks without exact interval/component labels; not estimated.
Related research questions
References
- [1] CharXiv
- [2] Evaluating Kimi K3 for Misalignment Risks. 2026-09-11. September 2026 ; §1.1.1 Figure 1 p.8