Acropolis

Muse Spark Open

MetaGrade A. Evidence grade A of A to D: strongest.Open weightsReleased 15 Oct 2026Within-domain variance
Compare with peers

Domain profile

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

  • Agentic (9 boards): 22 (scale 0 to 100)Agentic (9 boards)22
  • Coding (5 boards): 14 (scale 0 to 100)Coding (5 boards)14
  • Finance (4 boards): 20 (scale 0 to 100)Finance (4 boards)20
  • General text (1 boards): 59 (scale 0 to 100)General text (1 boards)59
  • Legal (3 boards): 31 (scale 0 to 100)Legal (3 boards)31
  • Science (2 boards): 67 (scale 0 to 100)Science (2 boards)67

Benchmark appearances

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

BenchmarkRankScoreCostStatusObserved
Legal Agent Benchmark (HLAB-Vals)

Vals AI benchmarks · xhigh

#2 / 38Tied #1
25%[17.9% to 32.1%]$13.63Standard15 Nov 2026
Finance Agent

Vals AI benchmarks · xhigh

#23 / 38
60.7%[56.6% to 64.7%]$2.76Standard15 Nov 2026
Legal Research

Vals AI benchmarks · xhigh

#26 / 38
38.5%[31.8% to 45.1%]$1.56Standard15 Nov 2026
Tax Agent

Vals AI benchmarks · xhigh

#28 / 38
26.3%[21.1% to 31.5%]$0.88Standard15 Nov 2026
Vibe Code Bench

Vals AI benchmarks · xhigh · OpenHands

#30 / 38
76.2%[69.7% to 82.7%]$1.66Standard15 Nov 2026
Code Migration

Vals AI benchmarks · xhigh

#33 / 38
25.1%[18.2% to 31.9%]$4.64Standard15 Nov 2026
Terminal-Bench 4

Vals AI benchmarks · xhigh

#33 / 38
6.6%[5.6% to 7.6%]$7.26Standard15 Nov 2026
Vals Index

Vals AI benchmarks · xhigh

#33 / 38
48.6%[46.4% to 50.7%]$4.72Standard15 Nov 2026
EMB

Vals AI benchmarks · xhigh

#34 / 38
62%[56.7% to 67.3%]$5.35Standard15 Nov 2026
LMArena WebDev

LMArena · Overall · max

#60 / 156
1,527[1,513 to 1,542]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev · max

#60 / 156
1,527[1,513 to 1,542]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev react · max

#60 / 140
1,525[1,506 to 1,544]UnknownStandard15 Nov 2026
GPQA diamond

Epoch AI Benchmarking Hub · xhigh

#72 / 338
88.6%[84.9% to 92.2%]UnknownStandard15 Oct 2026
LMArena WebDev

LMArena · Webdev HTML · max

#77 / 155
1,480[1,440 to 1,519]UnknownStandard15 Nov 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · xhigh

#113 / 299
57.5%UnknownStandardNot stated
ARC-AGI-2

Epoch AI Benchmarking Hub · high

#130 / 299
47.5%UnknownStandardNot stated
ARC-AGI-2

Epoch AI Benchmarking Hub · medium

#157 / 299
27.5%UnknownStandardNot stated
ARC-AGI-2

Epoch AI Benchmarking Hub · low

#180 / 299
12.2%UnknownStandardNot stated

Reasoning effort scaling

Same benchmark and harness measured across multiple reasoning effort settings.

  • ARC-AGI-2

    Epoch AI Benchmarking Hub

    Monotone. Evidence grade A of A to D: strongest.
    low

    12.2

    medium

    27.5

    high

    47.5

    xhigh

    57.5

Closest head-to-head rivals

Direct pairwise win rates for Muse Spark Open across shared benchmark boards.