AIM Consistency
Professional Work
Zeta's internal-consistency reading of the same 1,001 customer-report summaries scored by AIM Grounding: a judge reads all of a summary's statements together and decides whether any contradict one another, separately from whether the data supports them. The score is the share of summaries with no contradiction, reported as Zeta's mean judge score. Because it scores the same outputs as AIM Grounding, it is pooled in that factor group as a component.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Sol | 98.9 |
| GPT-6.1 Sol | 98.4 |
| GPT-6 Luna | 97.1 |
| Claude Sonnet 5.5 | 94.5 |
| GLM-5.3 | 92.8 |