University of Basel RISE · Research and Infrastructure SupportHumanities Data Benchmark

stepfun/step-5-preview

StepFunvia OpenRouter
average±4.8 (95%)
68.2
leaderboard rankof 114
≈63
benchmarksof 12
12
per 1,000 documentsmean over its benchmarks
$27
seconds per inputmean response time
104.2
Compare with other models Its row in the Matrix All its test runs

By task

  • Data correction

    61.4

    Best of 114 models with all its benchmarks: claude-opus-5-5, 82.7.

  • Document understanding

    71.8

    Best of 107 models with all its benchmarks: gpt-6-astra, 97.0.

  • Information extraction

    82.3

    Best of 107 models with all its benchmarks: grok-4.6, 90.8.

  • Transcription

    56.1

    Best of 114 models with all its benchmarks: gemini-4-argon, 96.2.

By benchmark

BenchmarkScoreThe best current run on each benchmark, on the benchmark's metric scaled to 0–100. ± is the 95% sampling margin, from resampling the run's inputs.RankRank among all current models on that benchmark; ≈ for models that cannot be told apart within sampling error.Per 1,000 docss / inputBest run
Bibliographic Dataother prompt 78.5±18.9 ≈91of 115 $25$0.12 run 35.24 T2054 · 24 Oct 2026prompt_v2.txt
Blacklist Cards 96.0±2.6 ≈14of 114 $9.20$0.30 run 30.17 T2061 · 24 Oct 2026prompt.txt
Book Advert XML Repairother prompt 29.1±9.5 ≈41of 119 $29$1.46 run 95.72 T2063 · 24 Oct 2026prompt_v2.txt
Business Letters 59.0±5.9 ≈24of 114 $13$0.76 run 34.43 T2055 · 24 Oct 2026prompt.txt
Company Listsother prompt 90.1±6.2 ≈6of 114 $18$0.27 run 44.28 T2062 · 24 Oct 2026prompt_v2.txt
Duty Rosters 77.7±2.7 ≈46of 107 $77$1.61 run 213.69 T2067 · 24 Oct 2026prompt.txt
Fraktur Adverts 37.2±44.7 ≈97of 114 $43$0.21 run 403.2 T2058 · 24 Oct 2026prompt_optimized.txt
General Meeting Minutes 86.4±9.2 ≈25of 108 $34$0.30 run 76.82 T2066 · 24 Oct 2026prompt.txt
Library Cardsother prompt 88.5±1.5 ≈15of 114 $8.90$2.34 run 20.05 T2059 · 24 Oct 2026prompt_year-string.txt
Magazine Pagesother prompt 44.1±13.7 ≈47of 115 $14$0.62 run 34.1 T2065 · 24 Oct 2026prompt_explicit-json.txt
Medieval Manuscripts 37.5±22.0 ≈95of 114 $39$0.47 run 210.0 T2060 · 24 Oct 2026prompt.txt
Personnel Cards 93.7±4.7 ≈58of 114 $15$0.90 run 52.12 T2064 · 24 Oct 2026prompt.txt

Price

FromInput, $ per million tokensOutput, $ per million tokensSource
24 Oct 20261.02.7archived price page

Every run

  1. T2067Duty Rosters24 Oct 202677.7$1.61
  2. T2066General Meeting Minutes24 Oct 202686.4$0.30
  3. T2065Magazine Pages24 Oct 202644.1$0.62
  4. T2064Personnel Cards24 Oct 202693.7$0.90
  5. T2063Book Advert XML Repair24 Oct 202629.1$1.46
  6. T2062Company Lists24 Oct 202690.1$0.27
  7. T2061Blacklist Cards24 Oct 202696.0$0.30
  8. T2060Medieval Manuscripts24 Oct 202637.5$0.47
  9. T2059Library Cards24 Oct 202688.5$2.34
  10. T2058Fraktur Adverts24 Oct 202637.2$0.21
  11. T2057Business Letters24 Oct 2026 · 32 signed letters60.0$0.39
  12. T2056Business Letters24 Oct 2026 · 25 unsigned letters53.0$0.31
  13. T2055Business Letters24 Oct 2026 · all 57 letters59.0$0.76
  14. T2054Bibliographic Data24 Oct 202678.5$0.12