Acropolis

GPT-6.1 Terra

OpenAIGrade A. Evidence grade A of A to D: strongest.ProprietaryReleased 26 Oct 2026Within-domain variance
Compare with peers

Domain profile

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

  • Agentic (12 boards): 68 (scale 0 to 100)Agentic (12 boards)68
  • Coding (7 boards): 78 (scale 0 to 100)Coding (7 boards)78
  • Finance (5 boards): 65 (scale 0 to 100)Finance (5 boards)65
  • Legal (3 boards): 5 (scale 0 to 100)Legal (3 boards)5
  • Science (1 boards): 95 (scale 0 to 100)Science (1 boards)95

Benchmark appearances

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

BenchmarkRankScoreCostStatusObserved
LMArena WebDev

LMArena · Webdev HTML · max

#7 / 155
1,736[1,710 to 1,761]UnknownStandard15 Nov 2026
EMB

Vals AI benchmarks · max

#8 / 38Tied #1
73.3%[68.9% to 77.7%]$0.56On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
Code Migration

Vals AI benchmarks · max

#9 / 38
72.2%[65.9% to 78.4%]$6.02On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
LMArena Agent

LMArena · Overall · max

#10 / 64
0.1[0.1 to 0.1]$0.87On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
GPQA diamond

Epoch AI Benchmarking Hub · max

#11 / 338Tied #1
94.3%[91.4% to 97.2%]UnknownStandard26 Oct 2026
Terminal-Bench 4

Vals AI benchmarks · max

#11 / 38
49.5%[45.9% to 53.1%]$2.16Standard15 Nov 2026
LMArena WebDev

LMArena · Overall · max

#13 / 156
1,681[1,670 to 1,691]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev · max

#13 / 156
1,681[1,670 to 1,691]UnknownStandard15 Nov 2026
Vending-Bench 2

Epoch AI Benchmarking Hub · Default

#13 / 72
$9,107.46UnknownStandardNot stated
Vals Index

Vals AI benchmarks · max

#14 / 38
60.5%[58.6% to 62.5%]$2.94Standard15 Nov 2026
Vibe Code Bench

Vals AI benchmarks · max · OpenHands

#17 / 38
84%[78.4% to 89.6%]$6.00Standard15 Nov 2026
LMArena Agent

LMArena · Overall · medium

#19 / 64
0[0 to 0.1]$0.16On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · max · Codex CLI

#21 / 145
40.7$3.88On frontier. Evidence grade A of A to D: strongest.26 Oct 2026
LMArena WebDev

LMArena · Webdev react · max

#21 / 140
1,669[1,656 to 1,682]UnknownStandard15 Nov 2026
Finance Agent

Vals AI benchmarks · max

#33 / 38
56%[51.9% to 60.1%]$1.92Standard15 Nov 2026
Tax Agent

Vals AI benchmarks · max

#34 / 38
17.1%[11.8% to 22.4%]$2.16Standard15 Nov 2026
Legal Research

Vals AI benchmarks · max

#36 / 38
26%[20% to 31.9%]$3.22Standard15 Nov 2026
Legal Agent Benchmark (HLAB-Vals)

Vals AI benchmarks · max

#38 / 38
1.7%[0.5% to 5.4%]$1.48Standard15 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · xhigh · Codex CLI

#123 / 145
11.3$0.75Standard26 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · high · Codex CLI

#127 / 145
9.7$0.56Standard26 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · medium · Codex CLI

#135 / 145
7.3$0.31Standard27 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · low · Codex CLI

#136 / 145
6$0.30Standard27 Oct 2026

Reasoning effort scaling

Same benchmark and harness measured across multiple reasoning effort settings.

  • Bug Hunt Bench

    Bug Hunt Bench · Codex CLI

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

    6.0

    $0.30

    medium

    7.3

    $0.31

    high

    9.7

    $0.56

    xhigh

    11.3

    $0.75

    max

    40.7

    $3.88

  • LMArena Agent

    LMArena

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

    0.0

    $0.16

    max

    0.1

    $0.87

Closest head-to-head rivals

Direct pairwise win rates for GPT-6.1 Terra across shared benchmark boards.