Atlas

Compare Models

Gemini 3.8 Flash vs Gemini 4 Argon

Gemini 4 Argon leads by 5.2 capability points (≈95% interval +2.9 to +7.6): a clear lead over Gemini 3.8 Flash · P(higher) ≈ 100%

Wins 257 of 306 shared benchmarks (32 lost, 17 tied)

Capability vs. relative cost
Gemini 3.8 FlashGemini 4 ArgonClosed 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 FlashvsGemini 4 Argoncapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins32 – 257 · 306 shared, 17 tied

Comparing matching setups only (32 benchmarks excluded) · suites counted once
← Gemini 3.8 FlashGemini 4 Argon →

Gemini 3.8 FlashvsGemini 4 Argonwins by categoryData Nov 15, 2026Data: Vals AI, Artificial Analysis, benchmarks.bio and 65 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Argon minus Gemini 3.8 Flash.

ARC-AGI-3 (Semi-Private)General QA35.097.3+62.3
FrontierMath Tier 4 (v2) - PrivateMath22.078.0+56.1
Concept Synth INDUCTION Challenge64 Round 1General QA14.165.6+51.6
ProofBench v1.1Math48.099.0+51.0
Concept Synth INDUCTION Challenge100 Round 1General QA9.057.0+48.0
IOI (Vals v2)Code56.9100.0+43.1
ArXivMath 08/2026 (Native Harnesses)Math40.482.5+42.1
RareDiseaseBench V1 (35 evaluations)Science2.943.8+41.0
Furniture Assembly BenchmarkMultimodal31.772.5+40.8
Harvey's Legal Agent Benchmark - Funds Asset ManagementLegal0.040.0+40.0
Terminal-Bench-Science 0.1Science10.548.6+38.1
Terminal-Bench 4.0Code19.757.6+37.9
ArXivMath 09/2026 (Native Harnesses)Math50.087.7+37.7
Terminal-Bench Science 0.1 (Vals)Science8.644.3+35.7
Terminal-Bench 4.0 (Artificial Analysis)Code19.755.1+35.4
CrossSearch (pass^3)Search29.064.0+35.0
FrontierSWE V2 Worst@5Code9.544.5+34.9
CyberBench v1.1 OverallCybersecurity43.877.9+34.1
FrontierML ($200 wallet)Agents5.739.5+33.8
Time Horizon Index: KSPAgents18.852.5+33.7
FrontierSWE V2 Best@5Code31.664.3+32.7
Code Migration v1.1Code38.670.8+32.2
Code MigrationCode36.568.2+31.6
ChartographyMultimodal40.971.6+30.7
BrowseComp Extended (Mercor held-out)Search41.572.0+30.5
Code2Video-AA EloCode8991259+29.2
Visa Agentic Commerce BenchmarkAgents53.382.2+28.9
Decision-Bench v1.1Professional Work26.954.5+27.6
EnigmaEvalGeneral QA37.264.6+27.4
BrokenArXiv 08/2026 (Native Harnesses)Math27.454.2+26.8
GameBench 2Games64.089.2+25.2
ScBench V1 Clean RerunsScience34.059.1+25.1
AA-3D v1.0 EloCode9741147+25.1
RiemannBenchMath51.276.0+24.8
Terminal-Bench 2.1Code81.357.7−23.6
MineBench — Bradley-Terry RatingGames18062195+23.6
TextQuests Game ProgressGames54.177.3+23.2
BrokenArXiv 09/2026 (Native Harnesses)Math26.749.7+23.0
Bug Hunt BenchCode17.138.7+21.6
AA-PersonalityBench Conviction indexChat & Writing84.863.8−21.2
SnorkelLegalLegal25.346.0+20.7
HiL-Bench (Human-in-Loop Benchmark)Agents41.562.0+20.5
AutomationBench 1.0.6 (Zapier)Agents29.750.1+20.4
ALE-BenchCode21822747+20.0
Harvey's Legal Agent Benchmark - Antitrust CompetitionLegal0.020.0+20.0
Harvey's Legal Agent Benchmark - Banking FinanceLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Capital MarketsLegal20.040.0+20.0
Harvey's Legal Agent Benchmark - Corporate M&ALegal0.020.0+20.0
Harvey's Legal Agent Benchmark - Data Privacy CybersecurityLegal0.020.0+20.0
Harvey's Legal Agent Benchmark - ImmigrationLegal0.020.0+20.0
Harvey's Legal Agent Benchmark - Intellectual PropertyLegal0.020.0+20.0
Harvey's Legal Agent Benchmark - International Trade SanctionsLegal0.020.0+20.0
Harvey's Legal Agent Benchmark - Structured Finance SecuritizationLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Trusts Estates Private ClientLegal20.040.0+20.0
GDP.xlsxProfessional Work18.338.3+20.0
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine13.333.3+20.0
Artificial Analysis Coding Agent Index v2.0Indexes33.353.2+20.0
Vals Science IndexIndexes35.855.4+19.6
Agent Red Teaming k=100 (CAIS)Safety29.810.2+19.6
Artificial Analysis Intelligence Index v5.0.1Indexes30.850.4+19.6
AA-PersonalityBench Warmth indexChat & Writing60.079.1+19.2
AA-LCR Extended (Mercor held-out)Search55.674.7+19.1
VariantBench V1 Clean RerunsScience31.450.3+18.9
Humanity's Last Exam DiamondGeneral QA34.353.2+18.9
SAGEMath35.153.6+18.6
Build-A-Chip-Bench (20-task index run)Agents0.118.6+18.5
TissueMicroscopyBench V1 (49 evaluations)Science3.421.8+18.4
VIALSMultimodal29.447.2+17.8
FrontierMath Tiers 1-3 (v2) - PrivateMath68.486.0+17.5
AA-PersonalityBench Organization indexChat & Writing73.356.2−17.3
Mystery Game PuzzlesGames47.064.0+17.0
HoneyBench v0.2 (hack rate)Safety80.563.6+16.9
LiveBench 2.0Agents67.384.1+16.8
AutomationBench-AA 1.0.6Agents60.977.5+16.6
SpatialViz-BenchMultimodal78.194.2+16.1
TxBench-Antibody-Development V1 (60 evaluations)Science15.031.1+16.1
GDPval Extended (Mercor held-out rubric score)Professional Work60.476.4+16.0
DAYJOB: FinanceFinance & Accounting2.818.8+16.0
HPCT Refusal (CAIS)Safety13.029.0+16.0
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal11.427.4+16.0
MKA (Artificial Analysis)Professional Work35.351.3+16.0
LMArena Agent Arena Net Improvement (Steering Burden)Agents2.310.7+15.9
TxBench Oligo Discovery V1 (113 evaluations)Science35.151.0+15.9
Legal Research BenchLegal38.954.8+15.9
Blueprint-Bench 2Multimodal0.40.5+15.8
BioSecBench-Surveillance V1 Clean RerunsScience44.159.8+15.7
MetagenomicsBench V1Science26.742.3+15.7
AA-Briefcase v1.1 EloAgents12041485+15.4
Ramp Accounting BenchFinance & Accounting45.761.1+15.4
RSI-Exam 0.1 (Full 88)Agents0.30.5+15.4
SWE-bench Multimodal (Vals v1, 480 Tasks)Code40.856.0+15.2
HieroglyphBenchMultimodal54.669.7+15.1
PLE (Physical-Engineering's Last Exam)Professional Work28.243.2+15.0
HLAB-Vals-1.1Legal12.127.1+15.0
CloudOpsBench v1.0Agents42.857.5+14.8
TxBench Antibody Discovery V1 (100 evaluations)Science49.063.7+14.7
APEX-Agents 1.1Professional Work72.987.4+14.5
Bio Capabilities Index (SecureBio)Indexes153174+14.5
MBCT (SecureBio)Science48.963.2+14.3
BioMysteryBench (Vals Terminus 2)Science62.276.3+14.1
Toloka Arena (v3)Agents59.073.0+14.0
NoCoT-Bench NCRI 15.2General QA126151+13.9
Physical-Engineering's Last Exam (hidden test)Professional Work27.241.1+13.9
Humanity's Sixth SenseMultimodal41.655.4+13.8
Finance Agent v2 - PrecedentsFinance & Accounting36.149.8+13.7
SAFE-Teen Relationships (no prompt)Safety41.728.6+13.0
LMArena Code Arena WebDevCode15831682+13.0
SnorkelRevOpsProfessional Work5.418.4+13.0
WeirdML v3Science7.420.3+12.9
APEX-SWECode36.449.2+12.8
SWE Atlas v1.1 - OverallCode57.370.1+12.8
APEX-Accounting 1.1Finance & Accounting58.871.5+12.7
SnorkelManufacturingProfessional Work7.119.5+12.4
Tax Agent BenchFinance & Accounting66.879.0+12.3
IntPhys 2Multimodal62.574.6+12.1
VCT v2 (SecureBio)Science45.257.3+12.1
DAYJOB: AccountingFinance & Accounting6.018.0+12.0
LAB-Bench ProtocolQA (SecureBio)Science84.095.9+11.9
HPCT (SecureBio)Science62.474.2+11.8
Artificial Analysis Intelligence Index v4.3.2Indexes40.952.6+11.6
SAFE-Teen 106: Relationships (no prompt)Safety40.128.5+11.6
VCT (SecureBio)Science46.958.4+11.5
Snorkel Digital Work IndexIndexes16.928.3+11.4
Dual-Arm Sim (task completion)Agents0.011.3+11.3
Remote Labor Index (RLI)Professional Work5.817.1+11.2
WitnessBench v2.0Games6.717.9+11.2
LegalReasonBenchLegal39.650.8+11.2
Westworld Finance Diligence BenchFinance & Accounting49.760.7+11.1
AutoResearchExam (hidden-test AUARC)Agents0.50.6+11.1
LMCAGeneral QA52.963.7+10.8
MedCodeHealth & Medicine48.158.8+10.7
Vals RSI Index v1.1 (four tasks)Agents19.930.6+10.6
LMArena Agent Arena Bash RecoveryAgents-0.7-21.4−10.4
BioSecBench-RespiratoryVirus-Surveillance V1 (110 evaluations)Science7.317.6+10.3
HANDBOOK.md Agents v1.1Professional Work26.536.6+10.1
TxBench Preclinical Pharmacology V1 Clean RerunsScience54.764.7+10.0
DAYJOB: HealthcareHealth & Medicine0.810.8+10.0
Snorkel SWE-bench CLICode6.316.3+10.0
Snorkel WorkplaceAgentsProfessional Work11.221.2+10.0
MissionReportProfessional Work54.364.1+9.8
SnorkelFinance 2.0Finance & Accounting5.415.1+9.7
LMArena Agent Arena Praise vs ComplaintAgents4.924.1+9.6
SpatialBench V1 Clean RerunsScience59.469.0+9.6
Humanity's Last ExamGeneral QA46.255.6+9.4
BenchBench-CloningScience9.218.6+9.4
BuseyBench SVGMultimodal6.67.6+9.3
τ³-Banking (Mercor run, bm25_grep retrieval)Agents48.858.1+9.3
Conceptual Reasoning IndexIndexes65.772.4+9.3
GBENCH One-shot Coding GScoreGames58.667.9+9.2
Snorkel Agentic Coding 2.0Code32.641.8+9.2
MysteryMechanismScience36.545.5+9.0
TreasuryQA BenchProfessional Work51.060.0+9.0
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting68.177.0+8.9
CritPt (avg@5)Science18.327.1+8.9
SpeciEval (Pooled English Runs)Safety89.981.3−8.6
ScBench-Long V1 Clean RerunsScience19.928.4+8.5
BioSecBench-Function V1 Clean RerunsScience36.044.4+8.5
GDPval-AA v2.1Professional Work14421634+8.4
FrontierPhysics v0.1Science0.08.3+8.3
DualEntry Accounting AI BenchmarkFinance & Accounting83.291.1+7.9
CritPt v2 (agentic, avg@5)Science65.172.9+7.7
LMArena Agent Arena Bash ReliabilityAgents9.9-5.4−7.6
PCT Political Inconsistency (CAIS)Safety46.338.7+7.6
EpiBench V1 Clean RerunsScience21.428.9+7.6
Artificial Analysis Progress Index v1Indexes320342+7.5
Web Clone BenchCode1.08.6+7.5
LLM Creative Story WritingChat & Writing0.10.8+7.4
LMArena Agent Arena Steering BurdenAgents-1.313.2+7.2
sudo L7Code11.718.9+7.2
FrontierCode 1.1 Extended - ScoreCode53.560.5+7.1
AML BenchFinance & Accounting57.164.1+7.0
Vals Legal IndexIndexes69.876.8+7.0
FrontierCode 2.0 (Artificial Analysis)Code18.725.7+7.0
ERQAMultimodal79.686.5+6.9
τ³-Banking (Fireworks, banking_knowledge)Agents51.458.2+6.9
Financial Coaching BenchFinance & Accounting66.373.1+6.9
SAFE-Teen 106: Health Advice (under-18 prompt)Safety37.530.8+6.8
SAFE-Teen Health Advice (under-18 prompt)Safety36.129.6+6.5
FrontierCode 1.1 Extended - Pass RateCode59.666.1+6.5
AA-Omniscience IndexIndexes29.642.4+6.4
Epoch Capabilities Index (ECI)Indexes159167+6.4
SimpleQA Verified (Epoch anti-abstention)General QA69.776.0+6.3
Bedside BenchHealth & Medicine86.592.7+6.2
Finance Agent v2 - All PassFinance & Accounting49.755.8+6.1
Finance Agent v2 - Disclosure AnalysisFinance & Accounting68.574.5+6.0
LMArena Agent Arena Confirmed SuccessAgents1.313.2+6.0
MedXpertQA MM Extended (Mercor held-out)Health & Medicine60.866.7+5.9
TextQuests Average Cumulative HarmSafety24.426.1−5.9
AA-PersonalityBench Expressiveness indexChat & Writing79.184.8+5.8
AA-PersonalityBench Imagination indexChat & Writing41.046.7+5.8
Finance Agent v2 - AdjustmentsFinance & Accounting54.860.5+5.7
Vals Finance IndexIndexes60.566.0+5.5
VibeVideoBench (overall quality, no video = 0)Code2.83.1+5.4
NoCoT-Bench NCKI (kspine v3)General QA121128+5.4
HealthBench ProfessionalHealth & Medicine59.064.2+5.2
SciCode 1.0.1 (Artificial Analysis)Code56.661.8+5.2
GDP.pdfProfessional Work23.228.4+5.2
HardGeoBench Country AccuracyMultimodal67.562.5−5.0
HardGeoBench <=100miMultimodal60.055.0−5.0
ComplexConstraintsChat & Writing48.453.3+4.9
Immigration BenchLegal63.468.1+4.7
SAFE-Teen 106 worst branch: Urgent Safety (under-18 prompt)Safety34.830.2+4.6
Machiavelli Deception (CAIS)Safety86.585.0+4.6
MindCube TinyMultimodal92.296.7+4.5
SAFE-Teen 106 worst branch: Health Advice (under-18 prompt)Safety48.143.6+4.5
Design Arena Overall Code CategoriesCode12991326+4.5
Public Benefits Bench v1.1Professional Work65.369.8+4.5
SWE-Bench Pro V2 — FullCode94.999.2+4.4
LMArena Text Arena EloChat & Writing14971522+4.3
AA-LCR v1.1Search84.079.7−4.3
VCT-Refusal (CAIS)Safety6.410.7+4.3
DeepSWE v1.1Code73.777.9+4.2
SnorkelUnderwrite 2.0Finance & Accounting19.123.3+4.2
Longform Creative WritingChat & Writing76.880.9+4.1
Eval Connections One-Shot with TrapsGames88.092.0+4.0
WCB (SecureBio)Science57.161.1+4.0
Harvey HLAB — Criteria Pass RateLegal90.294.0+3.9
AA-PersonalityBench Slop indexChat & Writing71.467.6−3.8
MMMU ProMultimodal84.588.3+3.8
SAFE-Teen 106 worst branch: Urgent Safety (no prompt)Safety56.452.6+3.8
HardGeoBench <=10miMultimodal46.342.5−3.7
Arena Alignment IndexSafety75.879.4+3.6
Finance Agent v2 - Earnings AnalysisFinance & Accounting81.484.8+3.4
NeuroscienceBench V1 (91 evaluations)Science15.819.1+3.4
ChessBenchGames73.370.0−3.3
BioTIER RefusalSafety83.987.2+3.3
BioSimBench 1.0Science15.318.4+3.1
Excel Modeling BenchmarkFinance & Accounting72.275.2+3.0
SAFE-Teen 106 worst branch: Relationships (no prompt)Safety62.665.7−3.0
micro1 PharmaBenchHealth & Medicine5.68.6+3.0
WellnessBenchSafety47.050.0+3.0
MedScribeHealth & Medicine84.587.4+2.9
HealthBenchHealth & Medicine57.160.0+2.9
SAFE-Teen Health Advice (no prompt)Safety59.156.2+2.9
SAFE-Teen Urgent Safety (under-18 prompt)Safety15.212.3+2.8
Finance Agent v2 - Financial ModelingFinance & Accounting32.629.8−2.8
SAFE-Teen 106: Urgent Safety (under-18 prompt)Safety14.912.1+2.8
LLM Chess EloGames15461609+2.6
AA-AnalystAgent (pass^5)Professional Work46.348.8+2.5
ChessBench EloGames15271485−2.5
Vibe Code Bench 1–100Code18.821.2+2.4
GDPval (Mercor 213-task rubric score)Professional Work83.585.9+2.4
MASKSafety58.661.0+2.4
DAYJOB: TaxFinance & Accounting0.83.2+2.4
SAFE-Teen Urgent Safety (no prompt)Safety31.033.3−2.3
SAFE-Teen 106: Health Advice (no prompt)Safety59.657.4+2.2
DistressBenchSafety70.768.6−2.1
WMDP-Bio (SecureBio)Science89.491.5+2.1
ARC-AGI-2General QA89.291.3+2.1
ChessBench CoherenceGames94.692.6−2.0
Audio MultiChallenge (Text Output)Multimodal60.458.4−2.0
Excel Modeling Benchmark v1.1Finance & Accounting72.974.8+2.0
AA-PersonalityBench Informativeness indexChat & Writing81.082.9+1.9
AA-PersonalityBench Transparency indexChat & Writing23.821.9−1.9
Finance Agent v2.1Finance & Accounting65.867.7+1.9
DTBenchGeneral QA95.797.6+1.9
SAFE-Teen 106 worst branch: Health Advice (no prompt)Safety69.070.8−1.9
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine15.613.9−1.7
SAFE-Teen 106: Urgent Safety (no prompt)Safety31.533.1−1.7
ProgramBenchCode1.02.5+1.5
FrontierMath ErdősMath0.01.5+1.5
Access to Justice BenchmarkLegal67.168.5+1.4
LegalBenchLegal87.088.3+1.3
BioSecBench-Refusal V1 Clean RerunsSafety47.448.7+1.3
RCM-Bench-1Health & Medicine90.291.5+1.3
SAGE v1.1Math88.289.5+1.2
Finance Agent v2 - Market AnalysisFinance & Accounting78.579.7+1.2
HardGeoBench Avg DistanceMultimodal843880−1.1
OTIS Mock AIME 2024-2025Math98.9100.0+1.1
Last Translation Benchmark (LTBv1-eval, blind)Chat & Writing39.840.9+1.1
DAXBench v2Code87.288.3+1.1
Finance Agent v2 - ComparablesFinance & Accounting52.051.0−1.0
SAFE-Teen Relationships (under-18 prompt)Safety4.03.0+1.0
Chess Puzzles (task version 2.0)Games58.059.0+1.0
ACCoRDGeneral QA78.277.3−0.9
ChessBench AccuracyGames81.181.8+0.8
SAFE-Teen 106: Relationships (under-18 prompt)Safety3.73.0+0.7
Creative Writing v3 EloChat & Writing17441758+0.7
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine48.449.1+0.7
CUA-benchAgents4.24.8+0.7
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting80.981.5+0.7
Professional Reasoning Benchmark - LegalLegal48.949.4+0.5
MacroscopeBenchCode70.770.3−0.4
Professional Reasoning Benchmark - FinanceFinance & Accounting49.549.9+0.4
Creative Writing v3 Rubric ScoreChat & Writing15.615.6−0.2
Prophet Arena Arena Score (Agentic, OpenCode)Forecasting86.086.0+0.1
Prophet Arena Arena Score (Fixed Context)Forecasting85.385.30.0
LMArena Agent Arena Tool HallucinationAgents0.40.50.0
SAFE-Teen 106 worst branch: Relationships (under-18 prompt)Safety14.014.00.0
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0
Harvey's Legal Agent Benchmark - Arbitration International Dispute ResolutionLegal0.00.00.0
Harvey's Legal Agent Benchmark - Bankruptcy RestructuringLegal0.00.00.0
Harvey's Legal Agent Benchmark - Corporate GovernanceLegal20.020.00.0
Harvey's Legal Agent Benchmark - Emerging Companies Venture CapitalLegal40.040.00.0
Harvey's Legal Agent Benchmark - Employment LaborLegal0.00.00.0
Harvey's Legal Agent Benchmark - Energy Natural ResourcesLegal20.020.00.0
Harvey's Legal Agent Benchmark - Environmental ESGLegal0.00.00.0
Harvey's Legal Agent Benchmark - Healthcare Life SciencesLegal20.020.00.0
Harvey's Legal Agent Benchmark - InsuranceLegal0.00.00.0
Harvey's Legal Agent Benchmark - Litigation Dispute ResolutionLegal0.00.00.0
Harvey's Legal Agent Benchmark - Real EstateLegal0.00.00.0
Harvey's Legal Agent Benchmark - TaxLegal0.00.00.0
Harvey's Legal Agent Benchmark - White Collar Defense InvestigationsLegal0.00.00.0
LAB-Bench CloningScenarios (SecureBio)Science100.0100.00.0
Next.js AI Agent EvaluationsAgents100.0100.00.0
SpatialBench-Long V1 Clean RerunsScience28.628.60.0

Gemini 3.8 FlashvsGemini 4 Argonnormalised score marginsData Nov 15, 2026Data: Vals AI, Artificial Analysis, benchmarks.bio and 65 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…