Atlas

Compare Models

GPT-6.1 Terra vs Kimi K3.1 Pro

GPT-6.1 Terra leads by 1.5 capability points (≈95% interval −1.0 to +3.9): statistically tied · P(higher) ≈ 88%

Kimi K3.1 Pro wins 27 of 54 shared benchmarks (25 lost, 2 tied)

Capability vs. relative cost
GPT-6.1 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-6.1 TerravsKimi K3.1 Procapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins25 – 27 · 54 shared, 2 tied

Comparing matching setups only (60 benchmarks excluded) · suites counted once
← GPT-6.1 TerraKimi K3.1 Pro →

GPT-6.1 TerravsKimi K3.1 Prowins by categoryData Nov 15, 2026Data: Vals AI, Epoch AI, Arena and 16 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-6.1 Terra.

Mystery Game PuzzlesGames74.042.0−32.0
FrontierMath Tier 4 (v2) - PrivateMath92.765.9−26.8
ALE-BenchCode28752132−26.3
Furniture Assembly BenchmarkMultimodal79.255.8−23.3
τ³-Banking (Mercor run, bm25_grep retrieval)Agents37.159.8+22.7
Terminal-Bench-Science 0.1Science58.338.1−20.2
GBENCH One-shot Coding GScoreGames77.658.2−19.4
Legal Research Bench HealthLegal56.773.3+16.7
GDPval-AA v2.1Professional Work14761846+16.3
NoCoT-Bench NCRI 15.2General QA130103−15.2
AA-Briefcase v1.1 EloAgents14011678+15.2
Legal Research Bench Civil LitigationLegal14.829.6+14.8
Terminal-Bench 4.0Code49.563.6+14.1
LMArena Agent Arena Net Improvement (Steering Burden)Agents12.04.9−13.4
AutomationBench-AA 1.0.6Agents60.273.3+13.1
FrontierMath Tiers 1-3 (v2) - PrivateMath90.978.2−12.6
Code Migration v1.1 (Vals Index v2.2 subset)Code72.259.7−12.5
Legal Research Bench Business/CommercialLegal16.327.9+11.6
Legal Research Bench Administrative/RegulatoryLegal45.556.1+10.6
Legal Research Bench CriminalLegal17.528.1+10.5
HLAB-Vals-1.1Legal1.711.7+10.0
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal4.513.8+9.3
Legal Research Bench ImmigrationLegal9.118.2+9.1
Legal Research Bench FamilyLegal4.513.6+9.1
FrontierMemory v1.1Agents46.755.6+8.9
Tax Agent BenchFinance & Accounting55.063.9+8.8
Design Arena Overall Code CategoriesCode13001353+8.8
Humanity's Last Exam DiamondGeneral QA47.339.4−7.9
DualEntry Accounting AI BenchmarkFinance & Accounting78.286.1+7.9
Legal Research Bench Constitutional/Civil RightsLegal23.731.6+7.9
DAXBench v2Code93.686.2−7.4
Legal Research Bench Weighted PassLegal78.771.4−7.3
LMArena Agent Arena Praise vs ComplaintAgents20.08.0−6.0
Finance Agent v2.1Finance & Accounting56.061.5+5.5
LMArena Agent Arena Steering BurdenAgents16.66.7−4.9
Epoch Capabilities Index (ECI)Indexes168162−4.9
SimpleQA Verified (Epoch anti-abstention)General QA57.362.2+4.9
NoCoT-Bench NCKI (kspine v3)General QA110116+4.6
LMArena Agent Arena Confirmed SuccessAgents9.72.7−3.5
APEX-Agents 1.1Professional Work79.375.9−3.4
BuseyBench SVGMultimodal7.36.9−3.4
Excel Modeling Benchmark v1.1Finance & Accounting73.370.0−3.3
LMArena Agent Arena Bash ReliabilityAgents13.47.2−3.1
LiveBench 2.0Agents79.776.7−2.9
Vals Finance IndexIndexes53.356.0+2.7
Visa Agentic Commerce BenchmarkAgents51.153.3+2.2
GameBench 2Games85.884.0−1.8
APEX-Accounting 1.1Finance & Accounting63.762.4−1.3
Vibe Code Bench v1.1Code84.084.9+0.9
MMMU ProMultimodal84.784.1−0.6
Prophet Arena Arena Score (Agentic, OpenCode)Forecasting85.786.1+0.4
Prophet Arena Arena Score (Fixed Context)Forecasting85.285.4+0.2
LMArena Agent Arena Tool HallucinationAgents1.01.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0

GPT-6.1 TerravsKimi K3.1 Pronormalised score marginsData Nov 15, 2026Data: Vals AI, Epoch AI, Arena and 16 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…