Atlas

Compare Models

Gemini 3.8 Flash vs Kimi K3.1 Pro

Kimi K3.1 Pro leads by 3.2 capability points (≈95% interval +0.9 to +5.4): a clear lead over Gemini 3.8 Flash · P(higher) ≈ 100%

Wins 71 of 104 shared benchmarks (31 lost, 2 tied)

Capability vs. relative cost
Gemini 3.8 FlashKimi 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 3.8 FlashvsKimi K3.1 Procapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins31 – 71 · 104 shared, 2 tied

Comparing matching setups only (54 benchmarks excluded) · suites counted once
← Gemini 3.8 FlashKimi K3.1 Pro →

Gemini 3.8 FlashvsKimi K3.1 Prowins by categoryData Nov 15, 2026Data: Vals AI, SecureBio, Epoch AI and 32 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 3.8 Flash.

Terminal-Bench 4.0Code19.763.6+43.9
FrontierMath Tier 4 (v2) - PrivateMath22.065.9+43.9
ProofBench v1.1Math48.091.0+43.0
BrokenArXiv 08/2026 (Native Harnesses)Math27.464.3+36.9
BrokenArXiv 09/2026 (Native Harnesses)Math26.763.5+36.8
Concept Synth INDUCTION Challenge64 Round 1General QA14.146.9+32.8
VoxelBench Text RatingGames17982216+29.8
AA-Briefcase v1.1 EloAgents12041678+26.0
Concept Synth INDUCTION Challenge100 Round 1General QA9.034.0+25.0
IOI (Vals v2)Code56.981.9+24.9
ArXivMath 08/2026 (Native Harnesses)Math40.464.9+24.6
Furniture Assembly BenchmarkMultimodal31.755.8+24.2
FrontierML ($200 wallet)Agents5.729.6+23.9
CloudOpsBench v1.0Agents42.819.1−23.7
Terminal-Bench Science 0.1 (Vals)Science8.631.4+22.9
micro1 PharmaBenchHealth & Medicine5.627.5+21.9
Creative Writing v3 EloChat & Writing17442173+21.7
Code Migration v1.1 (Vals Index v2.2 subset)Code38.259.7+21.5
GameBench 2Games64.084.0+20.0
GDPval-AA v2.1Professional Work14421846+17.8
Bug Hunt BenchCode17.134.3+17.1
Legal Research Bench HealthLegal56.773.3+16.7
Legal Research Bench Weighted PassLegal87.571.4−16.0
ArXivMath 09/2026 (Native Harnesses)Math50.064.2+14.2
Legal Research Bench Business/CommercialLegal41.927.9−14.0
LMArena Code Arena WebDevCode15831688+13.7
BioMysteryBench (Vals Terminus 2)Science62.275.6+13.3
Chess Puzzles (task version 2.0)Games58.045.0−13.0
NoCoT-Bench NCRI 15.2General QA126103−12.6
DAYJOB: Finance (October 2026 rescoring)Finance & Accounting4.517.0+12.5
AutomationBench-AA 1.0.6Agents60.973.3+12.4
Legal Research Bench CriminalLegal40.428.1−12.3
τ³-Banking (Fireworks, banking_knowledge)Agents51.463.3+11.9
MBCT (SecureBio)Science48.960.7+11.8
MineBench — Bradley-Terry RatingGames18061992+11.3
τ³-Banking (Mercor run, bm25_grep retrieval)Agents48.859.8+11.0
Legal Research Bench Constitutional/Civil RightsLegal42.131.6−10.5
MysteryMechanismScience36.526.1−10.4
FrontierMath Tiers 1-3 (v2) - PrivateMath68.478.2+9.8
LiveBench 2.0Agents67.376.7+9.4
Legal Research Bench Administrative/RegulatoryLegal47.056.1+9.1
Legal Research Bench ImmigrationLegal27.318.2−9.1
Design Arena Overall Code CategoriesCode12991353+9.0
HoneyBench v0.2 (hack rate)Safety80.571.8+8.7
Vals Science IndexIndexes35.844.4+8.6
LAB-Bench CloningScenarios (SecureBio)Science100.091.5−8.5
DAYJOB: AccountingFinance & Accounting6.013.6+7.6
EnterpriseBench: CoreCraftProfessional Work58.566.1+7.6
SimpleQA Verified (Epoch anti-abstention)General QA69.762.2−7.5
Vibe Code Bench v1.1Code78.784.9+6.2
sudo L7 (October 2026 rescoring)Code10.616.7+6.1
LAB-Bench ProtocolQA (SecureBio)Science84.078.1−5.9
Bio Capabilities Index (SecureBio)Indexes153161+5.7
WCB (SecureBio)Science57.151.5−5.6
Creative Writing v3 Rubric ScoreChat & Writing15.616.6+5.1
Mystery Game PuzzlesGames47.042.0−5.0
Humanity's Last ExamGeneral QA46.251.1+4.9
LMArena Agent Arena Net Improvement (Steering Burden)Agents2.34.9+4.9
LMCAGeneral QA52.957.6+4.7
Legal Research Bench FamilyLegal18.213.6−4.5
DeepSWE v1.1Code71.075.5+4.5
Vals Finance IndexIndexes60.556.0−4.4
Finance Agent v2.1Finance & Accounting65.861.5−4.3
LMArena Agent Arena Steering BurdenAgents-1.36.7+4.0
RiemannBench (November 2026 rescoring)Math56.860.8+4.0
APEX-Accounting 1.1Finance & Accounting58.862.4+3.6
ORCA-bench Hard RCA (Fireworks 324-task subset)Agents29.532.9+3.4
Harvey LAB: In-House Contracting (held-out)Legal6.33.0−3.3
BuseyBench SVGMultimodal6.66.9+3.1
Longform Creative WritingChat & Writing76.879.9+3.1
APEX-Agents 1.1Professional Work72.975.9+3.0
Humanity's Last Exam DiamondGeneral QA34.337.3+3.0
NoCoT-Bench NCKI (kspine v3)General QA121116−3.0
Conceptual Reasoning IndexIndexes65.767.8+3.0
Tax Agent BenchFinance & Accounting66.863.9−2.9
DualEntry Accounting AI BenchmarkFinance & Accounting83.286.1+2.9
Excel Modeling Benchmark v1.1Finance & Accounting72.970.0−2.9
Epoch Capabilities Index (ECI)Indexes159162+2.7
LMArena Text Arena EloChat & Writing14971483−2.5
VCT (SecureBio)Science46.949.4+2.5
Access to Justice BenchmarkLegal67.169.5+2.4
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal11.413.8+2.4
DTBenchGeneral QA95.793.3−2.4
ALE-BenchCode21822132−1.8
LAB-Bench LitQA2 (SecureBio)Science51.553.1+1.6
LMArena Agent Arena Praise vs ComplaintAgents4.98.0+1.5
LMArena Agent Arena Bash ReliabilityAgents9.97.2−1.3
OTIS Mock AIME 2024-2025Math98.9100.0+1.1
HPCT (SecureBio)Science62.463.5+1.1
DAXBench v2Code87.286.2−1.1
VCT v2 (SecureBio)Science45.246.2+1.0
Epiq LAER CorporateBenchProfessional Work77.678.4+0.9
LMArena Agent Arena Confirmed SuccessAgents1.32.7+0.7
ACCoRDGeneral QA78.277.5−0.7
GBENCH One-shot Coding GScoreGames58.658.2−0.4
HLAB-Vals-1.1Legal12.111.7−0.4
MMMU ProMultimodal84.584.1−0.4
WMDP-Bio (SecureBio)Science89.489.7+0.3
LMArena Agent Arena Tool HallucinationAgents0.41.0+0.3
Prophet Arena Arena Score (Fixed Context)Forecasting85.385.4+0.1
Political Science LLM BenchmarkProfessional Work68.768.6−0.1
Prophet Arena Arena Score (Agentic, OpenCode)Forecasting86.086.1+0.1
Legal Research Bench Civil LitigationLegal29.629.60.0
Visa Agentic Commerce BenchmarkAgents53.353.30.0

Gemini 3.8 FlashvsKimi K3.1 Pronormalised score marginsData Nov 15, 2026Data: Vals AI, SecureBio, Epoch AI and 32 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…