Atlas

Compare Models

GPT-5.6 Terra vs Kimi K3.1 Pro

Kimi K3.1 Pro leads by 3.2 capability points (≈95% interval +1.0 to +5.3): a clear lead over GPT-5.6 Terra · P(higher) ≈ 100%

Wins 57 of 79 shared benchmarks (18 lost, 4 tied)

Capability vs. relative cost
GPT-5.6 TerraKimi 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.

GPT-5.6 TerravsKimi K3.1 Procapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins18 – 57 · 79 shared, 4 tied

Comparing matching setups only (39 benchmarks excluded) · suites counted once
← GPT-5.6 TerraKimi K3.1 Pro →

GPT-5.6 TerravsKimi K3.1 Prowins by categoryData Nov 15, 2026Data: Vals AI, SecureBio, Epoch AI and 22 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 GPT-5.6 Terra.

Terminal-Bench 4.0Code22.763.6+40.9
GameBench 2Games53.584.0+30.6
VoxelBench Text RatingGames18942216+22.9
Terminal-Bench Science 0.1 (Vals)Science10.031.4+21.4
τ³-Banking (Mercor run, bm25_grep retrieval)Agents38.859.8+21.0
ZeroBench (Main questions, pass@5)Multimodal19.039.0+20.0
Design Arena Overall Code CategoriesCode12381353+19.1
AA-Briefcase v1.1 EloAgents13331678+18.9
DeepsecBench — Revalidated V1 (233 issues)Code19.137.5+18.4
SRE Bench — Fully SolvedCode9.527.1+17.6
ProofBench v1.1Math74.091.0+17.0
GDPval-AA v2.1Professional Work14611846+17.0
Creative Writing v3 EloChat & Writing18562173+16.0
ZeroBench Main Mean pass@1 (five samples)Multimodal14.229.8+15.6
DeepSWE v1.1Code60.275.5+15.3
Humanity's Last ExamGeneral QA35.951.1+15.2
BenchCAD - geometric overlapMultimodal78.292.7+14.5
DAYJOB: Finance (October 2026 rescoring)Finance & Accounting3.317.0+13.7
AutomationBench-AA 1.0.6Agents59.673.3+13.7
Legal Research Bench Weighted PassLegal85.171.4−13.6
LMArena Agent Arena Net Improvement (Steering Burden)Agents-1.84.9+12.7
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal2.313.8+11.5
Humanity's Last Exam DiamondGeneral QA28.839.4+10.6
Vibe Code Bench v1.1Code74.684.9+10.3
Concept Synth INDUCTION Challenge64 Round 1General QA37.546.9+9.4
Legal Research Bench Business/CommercialLegal37.227.9−9.3
Concept Synth INDUCTION Challenge100 Round 1General QA25.034.0+9.0
NoCoT-Bench NCKI (kspine v3)General QA104116+8.7
BuseyBench SVGMultimodal6.16.9+8.6
WCB (SecureBio)Science59.851.5−8.3
FrontierMath Tiers 1-3 (v2) - PrivateMath86.078.2−7.7
NoCoT-Bench NCRI 15.2General QA89.0103+7.6
Legal Research Bench CriminalLegal35.128.1−7.0
Mystery Game PuzzlesGames35.042.0+7.0
Legal Research Bench HealthLegal80.073.3−6.7
IOI (Vals v2)Code87.681.9−5.7
GBENCH One-shot Coding GScoreGames63.358.2−5.1
LMArena Agent Arena Confirmed SuccessAgents-7.22.7+5.0
LMArena Agent Arena Praise vs ComplaintAgents-1.98.0+4.9
FrontierMath Tier 4 (v2) - PrivateMath70.765.9−4.9
Bio Capabilities Index (SecureBio)Indexes155161+4.8
HPCT (SecureBio)Science68.263.5−4.7
Legal Research Bench FamilyLegal18.213.6−4.5
VCT v2 (SecureBio)Science41.946.2+4.3
APEX-Agents 1.1Professional Work71.875.9+4.1
MBCT (SecureBio)Science56.960.7+3.8
Creative Writing v3 Rubric ScoreChat & Writing15.916.6+3.6
LMArena Agent Arena Steering BurdenAgents-0.56.7+3.6
LMArena Agent Arena Bash ReliabilityAgents0.77.2+3.2
MMMU ProMultimodal80.783.9+3.2
CharXiv ReasoningMultimodal85.989.0+3.1
GPQA-Bio (SecureBio)Science85.382.2−3.1
Conceptual Reasoning IndexIndexes65.767.8+3.0
ACCoRDGeneral QA74.777.5+2.8
LMCAGeneral QA55.057.6+2.7
LMArena Text Arena EloChat & Writing14681483+2.6
Legal Research Bench Constitutional/Civil RightsLegal34.231.6−2.6
APEX-Accounting 1.1Finance & Accounting59.862.4+2.6
VCT (SecureBio)Science47.249.4+2.2
Chess Puzzles (task version 2.0)Games47.045.0−2.0
LAB-Bench LitQA2 (SecureBio)Science51.153.1+2.0
Longform Creative WritingChat & Writing78.079.9+1.9
LiveBench 2.0Agents75.076.7+1.8
Furniture Assembly BenchmarkMultimodal54.255.8+1.7
Legal Research Bench Administrative/RegulatoryLegal57.656.1−1.5
Tax Agent BenchFinance & Accounting65.263.9−1.3
LAB-Bench ProtocolQA (SecureBio)Science77.078.1+1.1
LAB-Bench CloningScenarios (SecureBio)Science92.591.5−1.0
Epoch Capabilities Index (ECI)Indexes161162+0.9
PubMedQA (SecureBio)Health & Medicine80.481.2+0.8
ALE-BenchCode21542132−0.8
MMLU-Bio (SecureBio)Science90.791.2+0.5
Political Science LLM BenchmarkProfessional Work68.368.6+0.3
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
LMArena Agent Arena Tool HallucinationAgents1.01.00.0
DTBenchGeneral QA93.393.30.0
Legal Research Bench Civil LitigationLegal29.629.60.0
Legal Research Bench ImmigrationLegal18.218.20.0
WMDP-Bio (SecureBio)Science89.789.70.0

GPT-5.6 TerravsKimi K3.1 Pronormalised score marginsData Nov 15, 2026Data: Vals AI, SecureBio, Epoch AI and 22 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…