Acropolis

GPT-5.4 Pro

OpenAIGrade C. Evidence grade C of A to D: thin.ProprietaryReleased 5 Mar 2026Consistent across benchmarks
Compare with peers

Domain profile

Median within-board percentile by capability domain across boards with sufficient peer overlap.

  • General text (2 boards): 85 (scale 0 to 100)General text (2 boards)85
  • Math (2 boards): 93 (scale 0 to 100)Math (2 boards)93
  • Science (3 boards): 89 (scale 0 to 100)Science (3 boards)89

Benchmark appearances

Best verified run per board, with confidence intervals and Pareto frontier status.

BenchmarkRankScoreCostStatusObserved
FrontierMath-2025-02-28-Private

Epoch AI Benchmarking Hub · xhigh

#4 / 101Tied #1
50%[44.3% to 55.7%]UnknownStandard6 Mar 2026
HLE

Epoch AI Benchmarking Hub · Default

#6 / 56
44.3%[40.5% to 48.1%]UnknownStandardNot stated
GPQA diamond

Epoch AI Benchmarking Hub · xhigh

#9 / 338Tied #1
94.6%[91.5% to 97.7%]UnknownStandard20 Mar 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · xhigh

#48 / 299
83.3%$16.41StandardNot stated
ARC-AGI-2

Epoch AI Benchmarking Hub · xhigh

#48 / 299
83.3%UnknownStandardNot stated

Closest head-to-head rivals

Direct pairwise win rates for GPT-5.4 Pro across shared benchmark boards.