Table 4. Education and gender equity gaps across all 4 modelsa.
| Model | Modal MAEb | EADc | EAD | |
|---|---|---|---|---|
| EAD | Perception | Information | Perception | Information |
| Education gap (low-high) | ||||
| Sonnet 4.5 | 0.88 | 0.03 | 0.44 | −0.02 |
| Opus 4.6 | 0.13 | 0.03 | −0.13 | −0.06 |
| GPT-5.2 | −0.13 | 0.19 | −0.06 | 0.04 |
| GPT-4.1 | 0.25 | 0.13 | −0.08 | 0.05 |
| Gender gap (male-female) | ||||
| Sonnet 4.5 | 0.13 | −0.03 | −0.05 | −0.02 |
| Opus 4.6 | 0 | 0.13 | −0.02 | 0.04 |
| GPT-5.2 | 0.38 | −0.06 | −0.06 | 0.01 |
| GPT-4.1 | 0.63 | 0 | −0.03 | 0 |
The upper block reports the education gap (edu-low minus edu-high); the lower block reports the gender gap. Positive values indicate poorer alignment for the low-education or male subgroup, respectively. Italicized values mark the largest absolute gap per metric column within each block.
MAE: mean absolute error.
EAD: expected absolute distance.