PolicyBench
MiniMax M3
MiniMax · AI alone, no tools
United States rank
#18 of 23
US exact
72.4%
United States
United States benchmark
Exact match
72.4%
Bounded score
79.1%
Parse rate
100.0%
Eligibility flags
937/984
Score by programhardest first
| Program | Exact | Within 1% | n |
|---|---|---|---|
| Federal tax before refundable credits | 47.0% | 50.0% | 100 |
| State tax before refundable credits | 52.0% | 53.0% | 100 |
| Payroll tax | 63.0% | 69.0% | 100 |
| Federal refundable credits | 79.0% | 79.0% | 100 |
| State refundable credits | 79.0% | 79.0% | 100 |
| SNAP | 80.0% | 80.0% | 100 |
| Person-level Medicaid eligibility | 84.2% | 84.2% | 177 |
| Self-employment tax | 93.0% | 95.0% | 100 |
| Person-level Head Start eligibility | 94.7% | 94.7% | 38 |
| SSI | 96.0% | 96.0% | 100 |
| Person-level Medicare eligibility | 97.2% | 97.2% | 177 |
| Person-level WIC eligibility | 97.2% | 97.2% | 177 |
| Person-level Early Head Start eligibility | 97.4% | 97.4% | 38 |
| Person-level CHIP eligibility | 97.7% | 97.7% | 177 |
| Free school meals eligibility | 99.0% | 99.0% | 100 |
| Reduced-price school meals eligibility | 99.0% | 99.0% | 100 |
| TANF | 99.0% | 99.0% | 100 |
| Local income tax | 100.0% | 100.0% | 100 |
Hardest casesworst misses on positive references
- State tax before refundable credits4180% offReference $7 · predicted $291Inspect household #023
- Federal tax before refundable credits1901% offReference $500 · predicted $9,999Inspect household #046
- Federal tax before refundable credits940% offReference $1,346 · predicted $14,000Inspect household #039
- State tax before refundable credits192% offReference $514 · predicted $1,500Inspect household #039
- Federal tax before refundable credits107% offReference $3,789 · predicted -$250Inspect household #033
- Federal tax before refundable credits100% offReference $2,883 · predicted $0Inspect household #000
Scores are from the frozen manuscript snapshot under the AI-alone condition: one structured response per household, no tools, graded against PolicyEngine reference outputs. See the leaderboard and paper for methodology.