Acropolis

Llama 3.2 90B Vision Instruct

MetaGrade B. Evidence grade B of A to D: good.Open weightsReleased 24 Sept 2024Consistent across benchmarks
Compare with peers

Domain profile

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

  • Science (1 boards): 20 (scale 0 to 100)Science (1 boards)20
  • Vision (2 boards): 12 (scale 0 to 100)Vision (2 boards)12

Benchmark appearances

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

BenchmarkRankScoreCostStatusObserved
LMArena Vision

LMArena · Chinese · Default

#127 / 135
952[908 to 995]UnknownStandard9 Nov 2026
LMArena Vision (style control)

LMArena · Chinese · Default

#127 / 135
1,004[961 to 1,048]UnknownStandard9 Nov 2026
LMArena Vision (style control)

LMArena · Overall · Default

#148 / 167
1,032[1,023 to 1,042]UnknownStandard9 Nov 2026
LMArena Vision

LMArena · Overall · Default

#149 / 167
1,000[990 to 1,009]UnknownStandard9 Nov 2026
LMArena Vision

LMArena · English · Default

#150 / 167
1,013[1,001 to 1,025]UnknownStandard9 Nov 2026
LMArena Vision (style control)

LMArena · English · Default

#151 / 167
1,039[1,027 to 1,051]UnknownStandard9 Nov 2026
GPQA diamond

Epoch AI Benchmarking Hub · Default

#289 / 338
41%[36.5% to 45.5%]UnknownStandard27 Jan 2025

Closest head-to-head rivals

Direct pairwise win rates for Llama 3.2 90B Vision Instruct across shared benchmark boards.