Skip to main content
PolicyBench

DeepSeek V4 Pro

DeepSeek · AI alone, no tools
Current board, snapshot 2026-09-30
United States rank
#36 of 46
US exact
80.8%
United States

United States benchmark

Exact match
80.8%
Bounded score
91.7%
Parse rate
100.0%
Eligibility flags
954/975

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 Early Head Start eligibility100.0%3.1%
  3. Person-level WIC eligibility99.4%0.3%
  4. Person-level Medicare eligibility99.4%11%
  5. Self-employment tax99.0%2.1%
  6. TANF99.0%0.3%
  7. Person-level CHIP eligibility98.3%0.2%
  8. Free school meals eligibility98.0%0.7%
  9. Reduced-price school meals eligibility98.0%0.1%
  10. SSI97.9%2.0%
  11. Person-level Head Start eligibility97.4%1.2%
  12. Person-level Medicaid eligibility93.6%30%
  13. Federal refundable credits86.7%3.7%
  14. SNAP84.9%4.1%
  15. State refundable credits81.6%0.6%
  16. Payroll tax70.5%16%
  17. State tax before refundable credits57.6%6.0%
  18. Federal tax before refundable credits50.6%19%
Table view (exact, within 1%, outputs), hardest first
ProgramExactWithin 1%n
Federal tax before refundable credits50.6%51.9%79
State tax before refundable credits57.6%57.6%85
Payroll tax70.5%74.7%95
State refundable credits81.6%81.6%98
SNAP84.9%84.9%93
Federal refundable credits86.7%87.8%98
Person-level Medicaid eligibility93.6%93.6%173
Person-level Head Start eligibility97.4%97.4%38
SSI97.9%99.0%97
Free school meals eligibility98.0%98.0%100
Reduced-price school meals eligibility98.0%98.0%100
Person-level CHIP eligibility98.3%98.3%177
Self-employment tax99.0%99.0%100
TANF99.0%99.0%100
Person-level Medicare eligibility99.4%99.4%172
Person-level WIC eligibility99.4%99.4%177
Local income tax100.0%100.0%100
Person-level Early Head Start eligibility100.0%100.0%38

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 7,527 rows whose legacy threshold score is below 1. That universe contains 7,523 of 7,523 exact-match misses and 4 exact hits. Another 2,072 rows with bounded score below 100 were not selected and have no audit annotation.