Acropolis

Haiku 5.5

AnthropicGrade A. Evidence grade A of A to D: strongest.ProprietaryReleased 7 Oct 2026Consistent across benchmarks
Compare with peers

Domain profile

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

  • Agentic (14 boards): 49 (scale 0 to 100)Agentic (14 boards)49
  • Coding (9 boards): 51 (scale 0 to 100)Coding (9 boards)51
  • Finance (5 boards): 43 (scale 0 to 100)Finance (5 boards)43
  • General text (4 boards): 68 (scale 0 to 100)General text (4 boards)68
  • Legal (3 boards): 51 (scale 0 to 100)Legal (3 boards)51
  • Math (1 boards): 73 (scale 0 to 100)Math (1 boards)73
  • Science (3 boards): 73 (scale 0 to 100)Science (3 boards)73

Benchmark appearances

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

BenchmarkRankScoreCostStatusObserved
Vibe Code Bench

Vals AI benchmarks · max · OpenHands

#4 / 38Tied #1
90.4%[87.1% to 93.8%]$6.07On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
Finance Agent

Vals AI benchmarks · max

#11 / 38Tied #1
65.4%[61.3% to 69.4%]$1.64Standard15 Nov 2026
HLE

Epoch AI Benchmarking Hub · Default

#13 / 56
30.1%[26.5% to 33.7%]UnknownStandardNot stated
FrontierCode

Epoch AI Benchmarking Hub · max · Claude Code

#15 / 51
46.4%UnknownStandardNot stated
Legal Research

Vals AI benchmarks · max

#19 / 38
43.8%[37% to 50.5%]$1.65Standard15 Nov 2026
Terminal-Bench 4

Vals AI benchmarks · max

#19 / 38
36.4%[31.8% to 40.9%]$3.12Standard15 Nov 2026
Vals Index

Vals AI benchmarks · max

#19 / 38
58.7%[56.2% to 61.2%]$3.17Standard15 Nov 2026
Code Migration

Vals AI benchmarks · max

#22 / 38
50.2%[42% to 58.4%]$6.13Standard15 Nov 2026
Tax Agent

Vals AI benchmarks · max

#22 / 38
31.9%[25.4% to 38.5%]$1.16Standard15 Nov 2026
CursorBench

CursorBench · max · Cursor

#24 / 91
48.4%$1.12On frontier. Evidence grade A of A to D: strongest.Not stated
EMB

Vals AI benchmarks · max

#31 / 38
65.5%[58.2% to 72.7%]$1.36Standard15 Nov 2026
Legal Agent Benchmark (HLAB-Vals)

Vals AI benchmarks · max

#33 / 38
4.2%[1.2% to 7.1%]$4.26Standard15 Nov 2026
LMArena Agent

LMArena · Overall · high

#35 / 64
-0[-0 to 0]$0.08Standard15 Nov 2026
LMArena WebDev

LMArena · Overall · high

#39 / 156
1,593[1,585 to 1,601]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev · high

#39 / 156
1,593[1,585 to 1,601]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev react · high

#39 / 140
1,615[1,605 to 1,625]UnknownStandard15 Nov 2026
CursorBench

CursorBench · xhigh · Cursor

#40 / 91
44.3%$0.56On frontier. Evidence grade A of A to D: strongest.Not stated
LMArena WebDev

LMArena · Webdev HTML · high

#44 / 155
1,564[1,544 to 1,585]UnknownStandard15 Nov 2026
CursorBench

CursorBench · high · Cursor

#45 / 91
42.3%$0.32On frontier. Evidence grade A of A to D: strongest.Not stated
Vending-Bench 2

Epoch AI Benchmarking Hub · Default

#49 / 72
$3,602.46UnknownStandardNot stated
CursorBench

CursorBench · medium · Cursor

#58 / 91
36.9%$0.17On frontier. Evidence grade A of A to D: strongest.Not stated
GPQA diamond

Epoch AI Benchmarking Hub · max

#62 / 338
89.6%[85.8% to 93.5%]UnknownStandard9 Oct 2026
LMArena Text (style control)

LMArena · German · high

#62 / 320
1,460[1,415 to 1,506]UnknownStandard11 Nov 2026
CursorBench

CursorBench · low · Cursor

#76 / 91
30.9%$0.08On frontier. Evidence grade A of A to D: strongest.Not stated
LMArena Text (style control)

LMArena · Math · high

#78 / 404
1,451[1,428 to 1,473]UnknownStandard11 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · max · Claude Code

#83 / 145
21.5$5.83Standard7 Oct 2026
LMArena Text (style control)

LMArena · Korean · high

#87 / 289
1,400[1,352 to 1,448]UnknownStandard11 Nov 2026
LMArena Text

LMArena · German · high

#89 / 320
1,429[1,384 to 1,474]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Math · high

#91 / 404
1,436[1,413 to 1,458]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Expert · high

#94 / 371
1,443[1,426 to 1,459]UnknownStandard11 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · xhigh · Claude Code

#95 / 145
18.3$4.80Standard7 Oct 2026
LMArena Text (style control)

LMArena · Japanese · high

#95 / 288
1,399[1,357 to 1,441]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Coding · high

#96 / 416
1,454[1,442 to 1,465]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Korean · high

#97 / 289
1,382[1,337 to 1,428]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry mathematical · high

#97 / 402
1,449[1,427 to 1,471]UnknownStandard11 Nov 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · xhigh

#98 / 299
61.3%UnknownStandardNot stated
LMArena Text (style control)

LMArena · Coding · high

#99 / 416
1,489[1,478 to 1,500]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry mathematical · high

#102 / 402
1,437[1,415 to 1,459]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Expert · high

#103 / 371
1,465[1,449 to 1,482]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · French · high

#103 / 301
1,455[1,427 to 1,482]UnknownStandard11 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · high · Claude Code

#104 / 145
16$3.56Standard7 Oct 2026
LMArena Text

LMArena · Instruction following · high

#104 / 421
1,411[1,402 to 1,421]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry software and IT services · high

#105 / 421
1,475[1,466 to 1,484]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Instruction following · high

#108 / 421
1,426[1,416 to 1,435]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Japanese · high

#111 / 288
1,374[1,331 to 1,416]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry software and IT services · high

#117 / 421
1,443[1,434 to 1,452]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Hard prompts · high

#117 / 421
1,449[1,442 to 1,457]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry business and management and financial operations · high

#120 / 414
1,430[1,417 to 1,443]UnknownStandard11 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · medium · Claude Code

#122 / 145
12$2.66Standard8 Oct 2026
LMArena Text

LMArena · French · high

#122 / 301
1,427[1,400 to 1,454]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry life and physical and social science · high

#122 / 419
1,446[1,433 to 1,460]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Chinese · high

#123 / 395
1,464[1,445 to 1,482]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Hard prompts English · high

#123 / 419
1,456[1,445 to 1,466]UnknownStandard11 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · low · Claude Code

#124 / 145
11$1.61Standard8 Oct 2026
LMArena Text (style control)

LMArena · Polish · high

#127 / 243
1,405[1,363 to 1,447]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Spanish · high

#129 / 304
1,414[1,388 to 1,440]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Exclude ties · high

#131 / 421
1,413[1,405 to 1,422]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · English · high

#132 / 421
1,433[1,424 to 1,442]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Non English · high

#133 / 421
1,406[1,398 to 1,413]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Overall · high

#133 / 421
1,421[1,415 to 1,427]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Hard prompts · high

#139 / 421
1,418[1,411 to 1,426]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry business and management and financial operations · high

#141 / 414
1,400[1,387 to 1,413]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Hard prompts English · high

#143 / 419
1,422[1,412 to 1,433]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry writing and literature and language · high

#146 / 420
1,391[1,381 to 1,401]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Longer query · high

#147 / 399
1,419[1,410 to 1,428]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Chinese · high

#149 / 395
1,433[1,415 to 1,452]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry entertainment and sports and media · high

#149 / 419
1,377[1,366 to 1,388]UnknownStandard11 Nov 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · high

#151 / 299
30.7%UnknownStandardNot stated
LMArena Text

LMArena · Spanish · high

#152 / 304
1,397[1,371 to 1,424]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Polish · high

#153 / 243
1,378[1,336 to 1,420]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Multi turn · high

#153 / 419
1,412[1,398 to 1,425]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry legal and government · high

#156 / 393
1,411[1,392 to 1,430]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Russian · high

#157 / 385
1,400[1,385 to 1,414]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Longer query · high

#159 / 399
1,395[1,386 to 1,404]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Exclude ties · high

#160 / 421
1,378[1,370 to 1,386]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry life and physical and social science · high

#160 / 419
1,411[1,398 to 1,425]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Non English · high

#160 / 421
1,385[1,377 to 1,392]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Overall · high

#160 / 421
1,398[1,392 to 1,404]UnknownStandard11 Nov 2026
LMArena Text

LMArena · English · high

#165 / 421
1,406[1,397 to 1,415]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry writing and literature and language · high

#166 / 420
1,369[1,359 to 1,380]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry entertainment and sports and media · high

#170 / 419
1,351[1,340 to 1,363]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Creative writing · high

#170 / 419
1,361[1,350 to 1,373]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Russian · high

#171 / 385
1,377[1,363 to 1,391]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry legal and government · high

#173 / 393
1,389[1,370 to 1,407]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Multi turn · high

#174 / 419
1,388[1,374 to 1,402]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry medicine and healthcare · high

#182 / 391
1,408[1,386 to 1,429]UnknownStandard11 Nov 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · medium

#185 / 299
9.4%UnknownStandardNot stated
LMArena Text

LMArena · Creative writing · high

#189 / 419
1,337[1,325 to 1,348]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry medicine and healthcare · high

#195 / 391
1,367[1,346 to 1,388]UnknownStandard11 Nov 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · low

#226 / 299
2.8%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

    2.8

    medium

    9.4

    high

    30.7

    xhigh

    61.3

  • Bug Hunt Bench

    Bug Hunt Bench · Claude Code

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

    11.0

    $1.61

    medium

    12.0

    $2.66

    high

    16.0

    $3.56

    xhigh

    18.3

    $4.80

    max

    21.5

    $5.83

  • CursorBench

    CursorBench · Cursor

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

    30.9

    $0.08

    medium

    36.9

    $0.17

    high

    42.3

    $0.32

    xhigh

    44.3

    $0.56

    max

    48.4

    $1.12

Closest head-to-head rivals

Direct pairwise win rates for Haiku 5.5 across shared benchmark boards.