Skip to main content
PolicyBench

GPT-5.4 nano

OpenAI · AI alone, no tools
Current board, snapshot 2026-10-14
United States rank
#50 of 50
US exact
65.4%
United States

United States benchmark

Exact match
65.4%
Bounded score
75.8%
Parse rate
100.0%
Eligibility flags
897/975

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.

  1. Local income tax100.0%0.3%
  2. Person-level CHIP eligibility100.0%0.2%
  3. Reduced-price school meals eligibility100.0%0.1%
  4. TANF99.0%0.3%
  5. SSI97.9%2.0%
  6. Person-level Early Head Start eligibility97.4%3.1%
  7. Person-level WIC eligibility96.0%0.3%
  8. Person-level Head Start eligibility94.7%1.2%
  9. Person-level Medicare eligibility93.0%11%
  10. Free school meals eligibility92.0%0.7%
  11. Self-employment tax91.0%2.1%
  12. SNAP86.0%4.1%
  13. Federal refundable credits86.0%3.7%
  14. State refundable credits81.8%0.6%
  15. Person-level Medicaid eligibility72.3%30%
  16. State tax before refundable credits55.9%6.0%
  17. Federal tax before refundable credits51.2%19%
  18. Payroll tax17.7%16%
Table view (exact, within 1%, outputs), hardest first
ProgramExactWithin 1%n
Payroll tax17.7%22.9%96
Federal tax before refundable credits51.2%52.4%84
State tax before refundable credits55.9%55.9%93
Person-level Medicaid eligibility72.3%72.3%173
State refundable credits81.8%81.8%99
Federal refundable credits86.0%86.0%100
SNAP86.0%86.0%93
Self-employment tax91.0%92.0%100
Free school meals eligibility92.0%92.0%100
Person-level Medicare eligibility93.0%93.0%172
Person-level Head Start eligibility94.7%94.7%38
Person-level WIC eligibility96.0%96.0%177
Person-level Early Head Start eligibility97.4%97.4%38
SSI97.9%97.9%97
TANF99.0%99.0%100
Local income tax100.0%100.0%98
Person-level CHIP eligibility100.0%100.0%177
Reduced-price school meals eligibility100.0%100.0%100

Hardest casesworst misses on positive references

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 8,765 rows whose legacy threshold score is below 1. That universe contains 8,761 of 8,761 exact-match misses and 4 exact hits. Another 2,329 rows with bounded score below 100 were not selected and have no audit annotation.