Acropolis

Kimi K2.7 Code

Moonshot AIGrade B. Evidence grade B of A to D: good.Open weightsReleased 12 Jun 2026Consistent across benchmarks
Compare with peers

Domain profile

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

  • Agentic (2 boards): 40 (scale 0 to 100)Agentic (2 boards)40
  • Coding (2 boards): 40 (scale 0 to 100)Coding (2 boards)40
  • Finance (1 boards): 49 (scale 0 to 100)Finance (1 boards)49
  • Science (1 boards): 71 (scale 0 to 100)Science (1 boards)71

Benchmark appearances

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

BenchmarkRankScoreCostStatusObserved
FrontierCode

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

#33 / 51
30.1%UnknownStandardNot stated
Vending-Bench 2

Epoch AI Benchmarking Hub · Default

#35 / 72
$5,082.94UnknownStandardNot stated
LMArena WebDev

LMArena · Image to webdev · Default

#41 / 64
1,512[1,491 to 1,532]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev react · Default

#76 / 140
1,475[1,464 to 1,485]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Overall · Default

#78 / 156
1,474[1,465 to 1,484]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev · Default

#78 / 156
1,474[1,465 to 1,484]UnknownStandard15 Nov 2026
GPQA diamond

Epoch AI Benchmarking Hub · Default

#82 / 338
87.9%[83.3% to 92.4%]UnknownStandard7 Aug 2026
LMArena WebDev

LMArena · Webdev HTML · Default

#86 / 155
1,453[1,429 to 1,478]UnknownStandard15 Nov 2026

Closest head-to-head rivals

Direct pairwise win rates for Kimi K2.7 Code across shared benchmark boards.