Atlas

Compare Models

Claude Opus 5 vs Kimi K3.1 Pro

Claude Opus 5 leads by 0.2 capability points (≈95% interval −2.1 to +2.6): statistically tied · P(higher) ≈ 58%

Wins 60 of 98 shared benchmarks (37 lost, 1 tied)

Capability vs. relative cost
Claude Opus 5Kimi K3.1 ProClosed weightsOpen weights

Each line runs through one model’s priced reasoning settings, cheapest to most expensive; grey points are other models’ priced settings in the same range. Results whose reasoning effort the source didn’t report are left out wherever a model has named effort levels. Capability and relative cost are the published estimates (relative cost is cost per task as a multiple of GPT-5 (high)); hover a setting, or use the arrow keys, to set it beside the other model’s same setting.

Claude Opus 5vsKimi K3.1 Procapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins60 – 37 · 98 shared, 1 tied

Comparing matching setups only (38 benchmarks excluded) · suites counted once
← Claude Opus 5Kimi K3.1 Pro →

Claude Opus 5vsKimi K3.1 Prowins by categoryData Nov 15, 2026Data: Vals AI, SecureBio, Arena and 31 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Kimi K3.1 Pro minus Claude Opus 5.

Legal Research Bench Business/CommercialLegal62.827.9−34.9
Legal Research Bench Civil LitigationLegal63.029.6−33.3
Legal Research Bench FamilyLegal40.913.6−27.3
Legal Research Bench Weighted PassLegal90.671.4−19.1
micro1 PharmaBenchHealth & Medicine9.727.5+17.8
ALE-BenchCode26222132−17.4
WCB (SecureBio)Science68.851.5−17.3
Mystery Game PuzzlesGames59.042.0−17.0
AutomationBench-AA 1.0.6Agents56.673.3+16.7
τ³-Banking (Mercor run, bm25_grep retrieval)Agents44.059.8+15.8
Legal Research Bench CriminalLegal43.928.1−15.8
HPCT (SecureBio)Science78.663.5−15.1
τ³-Banking (Fireworks, banking_knowledge)Agents49.163.3+14.2
Concept Synth INDUCTION Challenge64 Round 1General QA32.846.9+14.1
Legal Research Bench Constitutional/Civil RightsLegal44.731.6−13.2
ZeroBench (Main questions, pass@5)Multimodal26.039.0+13.0
Tax Agent BenchFinance & Accounting76.763.9−12.8
ZeroBench Main Mean pass@1 (five samples)Multimodal17.229.8+12.6
Bio Capabilities Index (SecureBio)Indexes178161−12.2
ORCA-bench Hard RCA (Fireworks 324-task subset)Agents44.332.9−11.4
MysteryMechanismScience37.426.1−11.3
DeepSWE v1.1Code65.075.5+10.5
Concept Synth INDUCTION Challenge100 Round 1General QA24.034.0+10.0
GBENCH One-shot Coding GScoreGames68.158.2−9.9
FrontierCode 2.0 - Pass RateCode31.422.0−9.4
Conceptual Reasoning IndexIndexes74.767.8−9.4
Legal Research Bench ImmigrationLegal27.318.2−9.1
Legal Research Bench Administrative/RegulatoryLegal65.256.1−9.1
Bug Hunt BenchCode25.734.3+8.6
LAB-Bench CloningScenarios (SecureBio)Science100.091.5−8.5
FrontierCode 2.0 - ScoreCode28.119.9−8.2
Terminal-Bench-Science 0.1Science30.038.1+8.1
Harvey LAB held-out set (Fireworks run)Legal7.415.4+8.1
ProofBench v1.1Math99.091.0−8.0
FrontierMath Tiers 1-3 (v2) - PrivateMath85.678.2−7.4
FrontierMath Tier 4 (v2) - PrivateMath73.265.9−7.3
LMCAGeneral QA64.557.6−6.8
DAYJOB: Finance (October 2026 rescoring)Finance & Accounting23.817.0−6.8
NoCoT-Bench NCRI 15.2General QA115103−6.7
BuseyBench SVGMultimodal6.36.9+6.4
Longform Creative WritingChat & Writing86.379.9−6.4
ACCoRDGeneral QA83.877.5−6.3
Chartography (with tools)Multimodal84.090.2+6.2
ProgramBench - almost solved rollout percentageCode37.031.5−5.5
Humanity's Last Exam DiamondGeneral QA44.839.4−5.4
LiveBench 2.0Agents82.176.7−5.4
Design Arena Overall Code CategoriesCode13211353+5.3
Humanity's Last ExamGeneral QA56.451.1−5.3
GDPval-AA v2.1Professional Work17291846+5.2
DeepsecBench — Revalidated V1 (233 issues)Code32.337.5+5.1
Furniture Assembly BenchmarkMultimodal60.855.8−5.0
DTBenchGeneral QA97.993.3−4.5
Epiq LAER CorporateBenchProfessional Work74.078.4+4.4
LMArena Agent Arena Net Improvement (Steering Burden)Agents7.24.9−4.4
ABC-Bench liquid handling (SecureBio)Science94.098.3+4.3
Terminal-Bench Science 0.1 (Vals)Science27.131.4+4.3
CharXiv Reasoning (with tools)Multimodal89.393.4+4.1
LMArena Agent Arena Confirmed SuccessAgents10.22.7−3.7
BioMysteryBench (Vals Terminus 2)Science79.375.6−3.7
Vals Science IndexIndexes47.944.4−3.6
Vibe Code Bench v1.1Code88.484.9−3.5
Legal Research Bench HealthLegal76.773.3−3.3
LAB-Bench ProtocolQA (SecureBio)Science81.178.1−3.0
Political Science LLM BenchmarkProfessional Work71.468.6−2.9
BenchCAD - geometric overlapMultimodal89.992.7+2.8
EnterpriseBench: CoreCraftProfessional Work68.766.1−2.6
Terminal-Bench 4.0Code50.352.8+2.5
IOI (Vals v2)Code84.381.9−2.4
LMArena Agent Arena Praise vs ComplaintAgents12.28.0−2.1
NoCoT-Bench NCKI (kspine v3)General QA119116−2.1
Chess Puzzles (task version 2.0)Games43.045.0+2.0
VCT (SecureBio)Science51.449.4−2.0
Epoch Capabilities Index (ECI)Indexes164162−2.0
Creative Writing v3 EloChat & Writing21352173+1.9
APEX-Accounting 1.1Finance & Accounting64.362.4−1.9
Creative Writing v3 Rubric ScoreChat & Writing16.316.6+1.8
LAB-Bench LitQA2 (SecureBio)Science51.453.1+1.7
ExploitGym - release-reported percentage, cohort unspecifiedCybersecurity22.120.7−1.4
BrowseCompSearch90.892.1+1.3
LMArena Agent Arena Bash ReliabilityAgents9.87.2−1.3
AA-Briefcase v1.1 EloAgents16541678+1.3
LMArena Agent Arena Steering BurdenAgents4.46.7+1.2
LMArena Text Arena EloChat & Writing14901483−1.2
OTIS Mock AIME 2024-2025Math98.9100.0+1.1
APEX-Agents 1.1Professional Work77.075.9−1.1
BigFinanceBench private holdout (final-answer accuracy)Finance & Accounting49.248.2−1.0
Harvey LAB: In-House Contracting (held-out)Legal2.03.0+1.0
VulnPR-100Code48.047.0−1.0
MineBench — Bradley-Terry RatingGames20081992−1.0
GameBench 2Games84.884.0−0.8
GPQA-Bio (SecureBio)Science82.782.2−0.5
MBCT (SecureBio)Science60.260.7+0.5
LMArena Code Arena WebDevCode16911688−0.4
SPU (Science Prediction Universe)Science0.20.2+0.3
MMMU ProMultimodal84.084.1+0.1
MMLU-Bio (SecureBio)Science91.191.2+0.1
LMArena Agent Arena Tool HallucinationAgents0.91.00.0
DAXBench v2Code86.286.20.0

Claude Opus 5vsKimi K3.1 Pronormalised score marginsData Nov 15, 2026Data: Vals AI, SecureBio, Arena and 31 othersVillage LLM Atlasatlas.village.dev

Win counts use each benchmark’s declared score direction. Represented suite slices and the derived Capability score are excluded from headline totals. Missing or incompatible setups are excluded, never counted as losses.
Loading Atlas data…