Atlas

Compare Models

Gemini 4 Argon vs Kimi K3.1 Pro

Gemini 4 Argon leads by 2.1 capability points (≈95% interval −0.3 to +4.5): statistically tied · P(higher) ≈ 95%

Wins 95 of 121 shared benchmarks (24 lost, 2 tied)

Capability vs. relative cost
Gemini 4 ArgonKimi 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.

Gemini 4 ArgonvsKimi K3.1 Procapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins95 – 24 · 121 shared, 2 tied

Comparing matching setups only (44 benchmarks excluded) · suites counted once
← Gemini 4 ArgonKimi K3.1 Pro →

Gemini 4 ArgonvsKimi K3.1 Prowins by categoryData Nov 15, 2026Data: Vals AI, SecureBio, Epoch AI and 39 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 Gemini 4 Argon.

CloudOpsBench v1.0Agents57.519.1−38.4
ArXivMath 10/2026 (Native Harnesses)Math83.045.7−37.2
Legal Research Bench ImmigrationLegal54.518.2−36.4
Visa Agentic Commerce BenchmarkAgents82.253.3−28.9
Legal Research Bench CriminalLegal56.128.1−28.1
NoCoT-Bench NCRI 15.2General QA151103−26.5
ALE-BenchCode28052132−23.8
ArXivMath 09/2026 (Native Harnesses)Math87.764.2−23.6
Concept Synth INDUCTION Challenge100 Round 1General QA57.034.0−23.0
Legal Research Bench FamilyLegal36.413.6−22.7
ArXivLean 09/2026Math39.617.0−22.6
Legal Research Bench Civil LitigationLegal51.929.6−22.2
Mystery Game PuzzlesGames64.042.0−22.0
ExploitGym - release-reported percentage, cohort unspecifiedCybersecurity42.020.7−21.3
Creative Writing v3 EloChat & Writing17582173+21.0
CWE-bench v2 Pass@1Cybersecurity65.045.3−19.7
Legal Research Bench Weighted PassLegal91.071.4−19.5
MysteryMechanismScience45.526.1−19.4
micro1 PharmaBenchHealth & Medicine8.627.5+18.9
Concept Synth INDUCTION Challenge64 Round 1General QA65.646.9−18.7
Legacy-Bench (Fireworks, avg@3)Code44.363.0+18.7
Legal Research Bench Business/CommercialLegal46.527.9−18.6
DAYJOB: Data ScienceProfessional Work23.34.7−18.6
BrokenArXiv 10/2026 (Native Harnesses)Math39.858.0+18.2
IOI (Vals v2)Code100.081.9−18.1
LAB-Bench ProtocolQA (SecureBio)Science95.978.1−17.8
ArXivMath 08/2026 (Native Harnesses)Math82.564.9−17.5
SRE Bench — Fully SolvedCode44.327.1−17.2
Furniture Assembly BenchmarkMultimodal72.555.8−16.7
Humanity's Last Exam DiamondGeneral QA53.237.3−15.9
Legal Research Bench Constitutional/Civil RightsLegal47.431.6−15.8
HLAB-Vals-1.1Legal27.111.7−15.4
Tax Agent BenchFinance & Accounting79.063.9−15.2
Code Migration v1.1 (Vals Index v2.2 subset)Code74.459.7−14.7
Chess Puzzles (task version 2.0)Games59.045.0−14.0
DAYJOB: LawLegal9.923.9+14.0
BrokenArXiv 09/2026 (Native Harnesses)Math49.763.5+13.8
SimpleQA Verified (Epoch anti-abstention)General QA76.062.2−13.8
RiemannBench (November 2026 rescoring)Math74.460.8−13.6
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal27.413.8−13.6
DAYJOB: Finance (October 2026 rescoring)Finance & Accounting30.517.0−13.5
Terminal-Bench Science 0.1 (Vals)Science44.331.4−12.9
MineBench — Bradley-Terry RatingGames21951992−12.3
FrontierMath Tier 4 (v2) - PrivateMath78.065.9−12.2
Legal Research Bench Administrative/RegulatoryLegal68.256.1−12.1
APEX-1 (Mercor, mean score)Professional Work81.169.3−11.8
APEX-Agents 1.1Professional Work87.475.9−11.5
FrontierCode 2.0 - Pass RateCode30.619.4−11.2
VCT v2 (SecureBio)Science57.346.2−11.1
LMArena Agent Arena Net Improvement (Steering Burden)Agents10.74.9−11.0
RWE-benchAgents12.01.0−11.0
FrontierBioMLScience36.825.8−11.0
Vals Science IndexIndexes55.444.4−11.0
HPCT (SecureBio)Science74.263.5−10.7
AA-Briefcase v1.1 EloAgents14851678+10.6
OpenWorldEngineer (Mean@8)Code24.113.6−10.5
BrokenArXiv 08/2026 (Native Harnesses)Math54.264.3+10.1
FrontierCode 2.0 - ScoreCode27.617.5−10.1
Vals Finance IndexIndexes66.056.0−10.0
FrontierML ($200 wallet)Agents39.529.6−9.9
GBENCH One-shot Coding GScoreGames67.958.2−9.7
WCB (SecureBio)Science61.151.5−9.6
GDPval-AA v2.1Professional Work16341846+9.4
APEX-Accounting 1.1Finance & Accounting71.562.4−9.1
VCT (SecureBio)Science58.449.4−9.0
FrontierMemory v1.1Agents64.455.6−8.9
Bio Capabilities Index (SecureBio)Indexes174161−8.8
LAB-Bench CloningScenarios (SecureBio)Science100.091.5−8.5
NoCoT-Bench NCKI (kspine v3)General QA128116−8.3
Terminal-Bench 4.0Code57.465.7+8.3
HoneyBench v0.2 (hack rate)Safety63.671.8−8.2
LMArena Agent Arena Praise vs ComplaintAgents24.18.0−8.1
ProofBench v1.1Math99.091.0−8.0
FrontierMath Tiers 1-3 (v2) - PrivateMath86.078.2−7.7
LiveBench 2.0Agents84.176.7−7.4
GPQA-Bio (SecureBio)Science89.382.2−7.1
LMArena Text Arena EloChat & Writing15221483−6.9
DeepFinanceFinance & Accounting41.434.6−6.8
BigFinanceBench private holdout (final-answer accuracy)Finance & Accounting54.748.2−6.5
Conceptual Reasoning IndexIndexes72.467.8−6.3
LMArena Agent Arena Bash ReliabilityAgents-5.47.2+6.3
BuseyBench SVGMultimodal7.66.9−6.2
Finance Agent v2.1Finance & Accounting67.761.5−6.2
LMCAGeneral QA63.757.6−6.1
Creative Writing v3 Rubric ScoreChat & Writing15.616.6+5.3
LMArena Agent Arena Confirmed SuccessAgents13.22.7−5.2
GameBench 2Games89.284.0−5.1
τ³-Banking (Fireworks, banking_knowledge)Agents58.263.3+5.1
DualEntry Accounting AI BenchmarkFinance & Accounting91.186.1−5.0
Excel Modeling Benchmark v1.1Finance & Accounting74.870.0−4.8
Humanity's Last ExamGeneral QA55.651.1−4.5
Design Arena Overall Code CategoriesCode13261353+4.5
Bug Hunt BenchCode38.734.3−4.4
DAYJOB: AccountingFinance & Accounting18.013.6−4.4
DTBenchGeneral QA97.693.3−4.3
MMMU ProMultimodal88.384.1−4.2
AutomationBench-AA 1.0.6Agents77.573.3−4.2
Epoch Capabilities Index (ECI)Indexes167162−3.6
Legal Research Bench HealthLegal76.773.3−3.3
Agents' Last Exam v1.1 (ALE v1.1)Agents40.243.4+3.3
LMArena Agent Arena Steering BurdenAgents13.26.7−3.2
Artificial Analysis Coding Agent Index v2.1Indexes45.548.6+3.1
MBCT (SecureBio)Science63.260.7−2.5
DAXBench v2Code88.386.2−2.1
WMDP-Bio (SecureBio)Science91.589.7−1.8
ErdosBenchMath3.03.0−1.7
ABC-Bench liquid handling (SecureBio)Science100.098.3−1.7
τ³-Banking (Mercor run, bm25_grep retrieval)Agents58.159.8+1.7
PubMedQA (SecureBio)Health & Medicine82.481.2−1.2
Longform Creative WritingChat & Writing80.979.9−1.0
Access to Justice BenchmarkLegal68.569.5+1.0
MMLU-Bio (SecureBio)Science92.191.2−0.9
BioMysteryBench (Vals Terminus 2)Science76.375.6−0.7
LMArena Code Arena WebDevCode16821688+0.7
sudo L7 (October 2026 rescoring)Code16.116.7+0.6
LMArena Agent Arena Tool HallucinationAgents0.51.0+0.2
ACCoRDGeneral QA77.377.5+0.2
Prophet Arena Arena Score (Fixed Context)Forecasting85.385.4+0.1
Prophet Arena Arena Score (Agentic, OpenCode)Forecasting86.086.10.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
Real-SWE 1.1Code39.239.20.0

Gemini 4 ArgonvsKimi K3.1 Pronormalised score marginsData Nov 15, 2026Data: Vals AI, SecureBio, Epoch AI and 39 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…