GPT-5.4 mini
United States benchmark
Exact match by programall 18 output groups, best first
Bar length is the unweighted exact-match rate on that program's outputs; the right column is the program's share of the headline weight, which follows household dollars, so the bars do not average to the headline.
- Local income tax100.0%0.3%
- Reduced-price school meals eligibility100.0%0.1%
- TANF99.0%0.3%
- SSI97.9%2.0%
- Person-level Medicare eligibility97.1%11%
- Free school meals eligibility95.0%0.7%
- Person-level Early Head Start eligibility92.1%3.1%
- Person-level CHIP eligibility92.1%0.2%
- Self-employment tax90.0%2.1%
- Person-level WIC eligibility86.4%0.3%
- Person-level Medicaid eligibility84.4%30%
- Person-level Head Start eligibility84.2%1.2%
- State refundable credits81.6%0.6%
- Federal refundable credits81.0%3.7%
- SNAP79.6%4.1%
- State tax before refundable credits60.2%6.0%
- Payroll tax56.3%16%
- Federal tax before refundable credits54.8%19%
Table view (exact, within 1%, outputs), hardest first
| Program | Exact | Within 1% | n |
|---|---|---|---|
| Federal tax before refundable credits | 54.8% | 54.8% | 84 |
| Payroll tax | 56.3% | 67.7% | 96 |
| State tax before refundable credits | 60.2% | 60.2% | 93 |
| SNAP | 79.6% | 79.6% | 93 |
| Federal refundable credits | 81.0% | 81.0% | 100 |
| State refundable credits | 81.6% | 81.6% | 98 |
| Person-level Head Start eligibility | 84.2% | 84.2% | 38 |
| Person-level Medicaid eligibility | 84.4% | 84.4% | 173 |
| Person-level WIC eligibility | 86.4% | 86.4% | 177 |
| Self-employment tax | 90.0% | 95.0% | 100 |
| Person-level CHIP eligibility | 92.1% | 92.1% | 177 |
| Person-level Early Head Start eligibility | 92.1% | 92.1% | 38 |
| Free school meals eligibility | 95.0% | 95.0% | 100 |
| Person-level Medicare eligibility | 97.1% | 97.1% | 172 |
| SSI | 97.9% | 97.9% | 97 |
| TANF | 99.0% | 99.0% | 100 |
| Local income tax | 100.0% | 100.0% | 98 |
| Reduced-price school meals eligibility | 100.0% | 100.0% | 100 |
Hardest casesworst misses on positive references
- SNAP635% offReference $288 · predicted $2,116Inspect household #030
- SNAP500% offReference $288 · predicted $1,728Inspect household #031
- SNAP307% offReference $288 · predicted $1,172Inspect household #108
- Federal tax before refundable credits100% offReference $2,906 · predicted $0Inspect household #000
- Federal tax before refundable credits100% offReference $22,401 · predicted $0Inspect household #003
- Federal tax before refundable credits100% offReference $3,708 · predicted $0Inspect household #007
Scores are from the frozen manuscript snapshot under the AI-alone condition: structured responses over the same household facts (whole-scenario or in output subsets, per the serving-configuration table), no tools, graded against PolicyEngine reference outputs. See the leaderboard and paper for methodology. The frozen US annotations cover 9,132 rows whose legacy threshold score is below 1. That universe contains 9,128 of 9,128 exact-match misses and 4 exact hits. Another 2,497 rows with bounded score below 100 were not selected and have no audit annotation.