Atlas

Compare Models

Gemini 4 Neon vs Kimi K3.1 Pro

Gemini 4 Neon leads by 5.6 capability points (≈95% interval +2.9 to +8.4): a clear lead over Kimi K3.1 Pro · P(higher) ≈ 100%

Wins 56 of 61 shared benchmarks (4 lost, 1 tied)

Capability vs. relative cost
Gemini 4 NeonKimi K3.1 ProClosed weightsOpen weights

Benchmark wins56 – 4 · 61 shared, 1 tied

Comparing matching setups only (32 benchmarks excluded) · suites counted once
← Gemini 4 NeonKimi K3.1 Pro →
Data: Vals AI, Surge AI, Arena and 10 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Kimi K3.1 Pro minus Gemini 4 Neon.

Mystery Game PuzzlesGames85.042.0−43.0
Humanity's Sixth SensePerception & media › Spatial & visual reasoning75.933.2−42.7
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine47.28.8−38.4
Chess Puzzles (task version 2.0)Games83.045.0−38.0
DAYJOB: AccountingProfessional › Finance, accounting & tax49.613.6−36.0
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax50.517.0−33.5
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning89.255.8−33.3
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge71.339.4−31.9
FrontierMath Tier 4 (v2) - PrivateMath97.665.9−31.7
Legal Research Bench Constitutional/Civil RightsProfessional › Legal63.231.6−31.6
Terminal-Bench Science 0.1 (Vals)Science › Multidisciplinary62.931.4−31.4
GDP.xlsxProfessional › General knowledge work43.114.0−29.1
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents18.43.9−28.3
Legal Research Bench Business/CommercialProfessional › Legal55.827.9−27.9
ExploitGym - release-reported percentage, cohort unspecifiedSoftware › Cybersecurity48.320.7−27.6
Legal Research Bench ImmigrationProfessional › Legal45.518.2−27.3
Bug Hunt BenchSoftware › Software engineering60.034.3−25.7
CWE-bench v2 Pass@1Software › Cybersecurity70.745.3−25.4
Legal Research Bench CriminalProfessional › Legal52.628.1−24.6
RiemannBench (November 2026 rescoring)Math84.060.8−23.2
Legal Research Bench FamilyProfessional › Legal36.413.6−22.7
sudo L7 (October 2026 rescoring)Software › Software engineering38.316.7−21.6
Legal Research Bench Weighted PassProfessional › Legal91.671.4−20.1
MysteryMechanismMath45.026.1−18.9
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents44.56.7−18.9
LiveBench 2.0 (2026-11-06 release)Composite indexes87.468.7−18.8
Legal Research Bench Civil LitigationProfessional › Legal48.129.6−18.5
Vals Science IndexScience › Multidisciplinary62.644.4−18.3
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering75.859.7−16.1
Tax Agent BenchProfessional › Finance, accounting & tax79.863.9−15.9
Legal Research Bench Administrative/RegulatoryProfessional › Legal71.256.1−15.2
APEX-Agents 1.1Professional › General knowledge work88.975.9−13.0
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge75.262.2−13.0
FrontierMath Tiers 1-3 (v2) - PrivateMath91.278.2−13.0
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal26.613.8−12.8
Vals Finance IndexProfessional › Finance, accounting & tax68.356.0−12.3
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations77.466.1−11.3
DAYJOB: TaxProfessional › Finance, accounting & tax12.81.6−11.2
LMArena Code Arena WebDevSoftware › Frontend & design17641684−10.6
APEX-1 (Mercor, mean score)Professional › General knowledge work79.469.3−10.1
Artificial Analysis Coding Agent Index v2.1Software › Software engineering58.648.6−10.0
Legal Research Bench HealthProfessional › Legal83.373.3−10.0
Vibe Code Bench v1.1Software › Frontend & design93.784.9−8.8
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents19.21.9−8.6
APEX-Accounting 1.1Professional › Finance, accounting & tax70.862.4−8.4
LMArena Agent Arena Steering BurdenTool & computer use › General agents21.24.5−8.3
HLAB-Vals-1.1Professional › Legal19.611.7−7.9
Finance Agent v2.1Professional › Finance, accounting & tax68.661.5−7.1
GDP.pdf (November 2026 rescoring)Professional › General knowledge work53.446.6−6.8
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax76.570.0−6.5
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-6.06.2+6.1
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations79.373.3−6.0
LMArena Text Arena EloLanguage & writing › General assistance15181487−5.6
BioMysteryBench (Vals Terminus 2)Science › Life sciences80.075.6−4.4
DAYJOB: Law (November 2026 release)Professional › Legal26.222.0−4.2
Terminal-Bench 4.0Software › Software engineering61.663.6+2.0
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work75.874.4−1.4
AA-Briefcase v1.1 EloProfessional › General knowledge work16991678−1.2
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations59.559.8+0.3
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.01.2+0.1
OTIS Mock AIME 2024-2025Math100.0100.00.0
Data: Vals AI, Surge AI, Arena and 10 othersatlas.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…