Acropolis

Kimi K3

Moonshot AIGrade A. Evidence grade A of A to D: strongest.Open weightsReleased 16 Jul 2026Consistent across benchmarks
Compare with peers

Domain profile

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

  • Agentic (13 boards): 53 (scale 0 to 100)Agentic (13 boards)53
  • Coding (8 boards): 44 (scale 0 to 100)Coding (8 boards)44
  • Finance (5 boards): 53 (scale 0 to 100)Finance (5 boards)53
  • General text (3 boards): 94 (scale 0 to 100)General text (3 boards)94
  • Legal (3 boards): 58 (scale 0 to 100)Legal (3 boards)58
  • Science (2 boards): 76 (scale 0 to 100)Science (2 boards)76

Benchmark appearances

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

BenchmarkRankScoreCostStatusObserved
LMArena Text (style control)

LMArena · Industry medicine and healthcare · max

#8 / 391Tied #1
1,515[1,503 to 1,527]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Japanese · max

#8 / 288
1,508[1,484 to 1,531]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry life and physical and social science · max

#10 / 419
1,518[1,510 to 1,527]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Japanese · max

#12 / 288
1,495[1,471 to 1,518]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry legal and government · max

#12 / 393
1,506[1,495 to 1,517]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Multi turn · max

#12 / 419
1,502[1,493 to 1,510]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Spanish · max

#12 / 304Tied #1
1,486[1,469 to 1,503]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Hard prompts · max

#13 / 421
1,516[1,511 to 1,521]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry mathematical · max

#13 / 402
1,511[1,498 to 1,525]UnknownStandard11 Nov 2026
Legal Research

Vals AI benchmarks · max

#13 / 38
46.2%[39.4% to 52.9%]$3.47Standard15 Nov 2026
LMArena Text

LMArena · Industry mathematical · max

#14 / 402
1,507[1,494 to 1,520]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Coding · max

#14 / 416
1,538[1,531 to 1,545]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Expert · max

#14 / 371
1,528[1,517 to 1,538]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · German · max

#14 / 320
1,498[1,476 to 1,520]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Math · max

#14 / 404
1,502[1,488 to 1,516]UnknownStandard11 Nov 2026
EMB

Vals AI benchmarks · max

#15 / 38Tied #1
71.3%[66.8% to 75.7%]$3.07Standard15 Nov 2026
LMArena Text

LMArena · Industry legal and government · max

#15 / 393
1,498[1,487 to 1,509]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Chinese · max

#15 / 395
1,540[1,529 to 1,552]UnknownStandard11 Nov 2026
Vibe Code Bench

Vals AI benchmarks · max · OpenHands

#15 / 38
85%[79.6% to 90.4%]$10.01Standard15 Nov 2026
FrontierCode

Epoch AI Benchmarking Hub · none · Mini-SWE-Agent

#16 / 51
44.2%UnknownStandardNot stated
LMArena Text

LMArena · Multi turn · max

#16 / 419
1,489[1,481 to 1,497]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Hard prompts English · max

#16 / 419
1,515[1,508 to 1,521]UnknownStandard11 Nov 2026
Legal Agent Benchmark (HLAB-Vals)

Vals AI benchmarks · max

#16 / 38
12.9%[7.4% to 18.4%]$4.14Standard15 Nov 2026
Tax Agent

Vals AI benchmarks · max

#16 / 38
33.4%[26.7% to 40%]$2.79Standard15 Nov 2026
LMArena Text

LMArena · Expert · max

#17 / 371
1,515[1,505 to 1,525]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry business and management and financial operations · max

#17 / 414
1,488[1,481 to 1,496]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Instruction following · max

#17 / 421
1,489[1,483 to 1,495]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Longer query · max

#17 / 399
1,500[1,494 to 1,506]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Coding · max

#18 / 416
1,506[1,500 to 1,513]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry life and physical and social science · max

#18 / 419
1,502[1,494 to 1,510]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Instruction following · max

#18 / 421
1,484[1,478 to 1,490]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Longer query · max

#18 / 399
1,490[1,484 to 1,496]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · English · max

#18 / 421
1,493[1,487 to 1,499]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Hard prompts · max

#19 / 421
1,494[1,489 to 1,499]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Math · max

#19 / 404
1,496[1,482 to 1,510]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Korean · max

#19 / 289Tied #1
1,457[1,435 to 1,479]UnknownStandard11 Nov 2026
LMArena WebDev

LMArena · Overall · max

#19 / 156
1,658[1,652 to 1,663]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev · max

#19 / 156
1,658[1,652 to 1,663]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev react · max

#19 / 140
1,673[1,666 to 1,679]UnknownStandard15 Nov 2026
LMArena Text (style control)

LMArena · Industry software and IT services · max

#20 / 421
1,523[1,517 to 1,528]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Overall · max

#20 / 421
1,487[1,483 to 1,492]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry medicine and healthcare · max

#21 / 391
1,488[1,476 to 1,500]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Spanish · max

#21 / 304
1,477[1,460 to 1,494]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Russian · max

#21 / 385
1,493[1,485 to 1,502]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Hard prompts English · max

#22 / 419
1,493[1,487 to 1,500]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry software and IT services · max

#22 / 421
1,497[1,492 to 1,503]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Exclude ties · max

#22 / 421
1,501[1,496 to 1,507]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Non English · max

#22 / 421
1,477[1,472 to 1,482]UnknownStandard11 Nov 2026
LMArena WebDev

LMArena · Image to webdev · max

#22 / 64
1,585[1,576 to 1,595]UnknownStandard15 Nov 2026
LMArena Text (style control)

LMArena · French · max

#23 / 301
1,501[1,485 to 1,518]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry writing and literature and language · max

#23 / 420
1,474[1,468 to 1,481]UnknownStandard11 Nov 2026
LMArena WebDev

LMArena · Webdev HTML · max

#23 / 155
1,640[1,628 to 1,652]UnknownStandard15 Nov 2026
LMArena Text

LMArena · English · max

#24 / 421
1,479[1,474 to 1,485]UnknownStandard11 Nov 2026
LMArena Text

LMArena · German · max

#24 / 320
1,483[1,461 to 1,504]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry business and management and financial operations · max

#25 / 414
1,469[1,462 to 1,477]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Industry entertainment and sports and media · max

#25 / 419
1,459[1,452 to 1,466]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Polish · max

#25 / 243
1,489[1,467 to 1,511]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Exclude ties · max

#26 / 421
1,482[1,476 to 1,487]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry writing and literature and language · max

#26 / 420
1,468[1,462 to 1,475]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Russian · max

#26 / 385
1,481[1,472 to 1,489]UnknownStandard11 Nov 2026
Finance Agent

Vals AI benchmarks · max

#27 / 38
60%[55.9% to 64.1%]$1.88Standard15 Nov 2026
LMArena Text

LMArena · Chinese · max

#27 / 395
1,527[1,516 to 1,538]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Korean · max

#27 / 289
1,444[1,422 to 1,466]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Industry entertainment and sports and media · max

#28 / 419
1,450[1,443 to 1,456]UnknownStandard11 Nov 2026
LMArena Text

LMArena · Overall · max

#28 / 421
1,475[1,470 to 1,479]UnknownStandard11 Nov 2026
Terminal-Bench 4

Vals AI benchmarks · max

#28 / 38
17.2%[16.2% to 18.2%]$12.48Standard15 Nov 2026
GPQA diamond

Epoch AI Benchmarking Hub · max

#29 / 338Tied #1
93.1%[90.2% to 96%]UnknownStandard16 Jul 2026
LMArena Text

LMArena · Non English · max

#29 / 421
1,465[1,460 to 1,470]UnknownStandard11 Nov 2026
Vals Index

Vals AI benchmarks · max

#30 / 38
53%[51% to 54.9%]$7.17Standard15 Nov 2026
LMArena Text

LMArena · Polish · max

#31 / 243
1,472[1,450 to 1,494]UnknownStandard11 Nov 2026
LMArena Text (style control)

LMArena · Creative writing · max

#31 / 419
1,459[1,452 to 1,467]UnknownStandard11 Nov 2026
Vending-Bench 2

Epoch AI Benchmarking Hub · Default

#32 / 72
$5,165.04UnknownStandardNot stated
LMArena Agent

LMArena · Overall · max

#33 / 64
-0[-0 to 0]$0.89Standard15 Nov 2026
LMArena Text

LMArena · French · max

#33 / 301
1,482[1,465 to 1,498]UnknownStandard11 Nov 2026
Code Migration

Vals AI benchmarks · max

#34 / 38
22.7%[15.7% to 29.6%]$19.51Standard15 Nov 2026
LMArena Text

LMArena · Creative writing · max

#34 / 419
1,453[1,445 to 1,460]UnknownStandard11 Nov 2026
GPQA diamond

Epoch AI Benchmarking Hub · high

#37 / 338
91.9%[88.1% to 95.7%]UnknownStandard7 Aug 2026
Bug Hunt Bench

Bug Hunt Bench · Claude Code

#84 / 145
21$25.27Standard26 Jul 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · max

#103 / 299
60.4%$1.59StandardNot stated
GPQA diamond

Epoch AI Benchmarking Hub · low

#111 / 338
84.8%[79.8% to 89.9%]UnknownStandard7 Aug 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · high

#117 / 299
55%$0.95StandardNot stated
ARC-AGI-2

Epoch AI Benchmarking Hub · low

#179 / 299
12.4%$0.25StandardNot 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

    12.4

    $0.25

    high

    55.0

    $0.95

    max

    60.4

    $1.59

  • GPQA diamond

    Epoch AI Benchmarking Hub

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

    84.8

    high

    91.9

    max

    93.1

Closest head-to-head rivals

Direct pairwise win rates for Kimi K3 across shared benchmark boards.