Atlas

Compare Models

DeepSeek V4.1 Flash vs Gemini 3.8 Flash

Gemini 3.8 Flash leads by 1.3 capability points (≈95% interval −0.8 to +3.4): statistically tied · P(higher) ≈ 89%

Wins 136 of 243 shared benchmarks (104 lost, 3 tied)

Capability vs. relative cost
DeepSeek V4.1 FlashGemini 3.8 FlashClosed 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.

DeepSeek V4.1 FlashvsGemini 3.8 Flashcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins104 – 136 · 243 shared, 3 tied

Comparing matching setups only (25 benchmarks excluded) · suites counted once
← DeepSeek V4.1 FlashGemini 3.8 Flash →

DeepSeek V4.1 FlashvsGemini 3.8 Flashwins by categoryData Nov 15, 2026Data: Vals AI, Artificial Analysis, Arena and 52 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 3.8 Flash minus DeepSeek V4.1 Flash.

Design Creativity BenchCode43.660.6+54.0
RSI Arena Stage 1 human ratingAgents887981+45.2
AA-PersonalityBench Transparency indexChat & Writing60.023.8−36.5
BioSecBench-Refusal V1 Clean RerunsSafety11.347.4+36.1
AA-PersonalityBench Slop indexChat & Writing37.271.4+34.6
ARC-AGI-3 (Semi-Private)General QA3.235.0+31.9
BlenderWorldCode65.734.2−31.5
Surface Evolver BenchAgents46.376.9+30.6
CyberBench v1.1 OverallCybersecurity73.743.8−29.9
SAFE-Teen 106: Health Advice (no prompt)Safety30.259.6−29.4
ChartographyMultimodal11.640.9+29.3
SWE-bench Verified Extended (Mercor held-out)Code82.754.0−28.7
BrokenArXiv 09/2026 (Native Harnesses)Math54.726.7−28.0
SAFE-Teen 106 worst branch: Health Advice (no prompt)Safety41.169.0−27.9
BrokenArXiv 08/2026 (Native Harnesses)Math54.827.4−27.4
RiemannBenchMath24.051.2+27.2
AA-PersonalityBench Organization indexChat & Writing10073.3−26.9
SimpleQA Verified (Epoch anti-abstention)General QA42.969.7+26.8
BuseyBench SVGMultimodal4.16.6+25.4
SAFE-Teen 106 worst branch: Urgent Safety (no prompt)Safety31.456.4−25.0
P2NPCBench (NP-complete, pass@1)Code39.362.6+23.3
APEX-Agents 1.1Professional Work50.572.9+22.4
MedXpertQA MM (200-question set)Health & Medicine63.584.7+21.2
CloudOpsBench v1.0Agents21.842.8+21.0
ScBench V1 Clean RerunsScience54.734.0−20.7
SkillsBench without SkillsAgents61.741.0−20.6
Code Migration v1.1Code58.938.6−20.3
SAFE-Teen 106: Urgent Safety (no prompt)Safety11.331.5−20.1
CadQueryEval (Corrected Geometry)Code76.096.0+20.0
MacroscopeBenchCode70.751.3−19.4
AA-PersonalityBench Warmth indexChat & Writing41.060.0+19.2
ArXivMath 04/2026Math45.064.2+19.2
LLM Chess EloGames10871546+18.8
BrokenArXiv 06/2026Math39.420.8−18.5
AA-Omniscience IndexIndexes-5.329.6+17.4
AA-PersonalityBench Expressiveness indexChat & Writing61.979.1+17.3
EnterpriseBench: CoreCraftProfessional Work41.558.5+17.0
VIALSMultimodal12.629.4+16.8
IOI (Vals v2)Code40.356.9+16.7
DeepsecBench-AA (Artificial Analysis)Cybersecurity31.214.5−16.7
ARC-AGI-2General QA72.989.2+16.3
OpenDS-WorldSearch20.536.5+16.0
LMArena Agent Arena Chat Net Improvement (Steering Burden)Agents1.68.2+16.0
SUPERChemScience56.772.6+15.9
CWE-Bench-AA (Artificial Analysis)Cybersecurity68.352.5−15.8
SimpleBenchGeneral QA66.782.4+15.7
Artificial Analysis Cyber Index v1Indexes40.825.1−15.7
τ³-Banking (Fireworks, banking_knowledge)Agents35.751.4+15.6
APEX-SWECode51.936.4−15.5
MysteryMechanismScience21.236.5+15.3
AA-LCR Extended (Mercor held-out)Search70.855.6−15.2
Finance Agent v2 - Market AnalysisFinance & Accounting63.478.5+15.1
Eval Connections One-Shot with TrapsGames73.088.0+15.0
Excel Modeling BenchmarkFinance & Accounting57.272.2+15.0
CyberGym-E2E-AA (Artificial Analysis)Cybersecurity22.98.4−14.5
SAFE-Teen 106: Health Advice (under-18 prompt)Safety23.637.5−13.9
SAFE-Teen 106: Relationships (no prompt)Safety27.140.1−13.0
AA-Briefcase EloAgents14241202−12.9
SWE Atlas - Codebase QnACode53.540.6−12.9
SAGEMath47.935.1−12.8
SheetBench-50Finance & Accounting58.445.6−12.8
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine60.948.4−12.5
Code2Video-AA EloCode1053899−12.5
ALE-BenchCode18292182+12.5
SWE-bench MultilingualCode98.285.8−12.4
MedXpertQA MM Extended (Mercor held-out)Health & Medicine48.460.8+12.4
SkillsBench with SkillsAgents69.858.0−11.8
DAXBench v2Code75.587.2+11.7
AA-PersonalityBench Conviction indexChat & Writing96.284.8−11.5
SAFE-Teen 106 worst branch: Relationships (no prompt)Safety51.162.6−11.5
MMMU-Pro Extended (Mercor held-out)Multimodal25.036.4+11.4
MetagenomicsBench V1Science38.026.7−11.3
MetagenomicsBench V0Science38.026.7−11.3
AA-AnalystAgent (pass^5)Professional Work35.046.3+11.3
SAFE-Teen 106 worst branch: Health Advice (under-18 prompt)Safety37.048.1−11.1
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting69.880.9+11.0
τ³-Banking (Mercor run, bm25_grep retrieval)Agents37.848.8+11.0
LiveBench 2.0Agents78.067.3−10.7
GBENCH One-shot Coding GScoreGames48.058.6+10.7
Creative Writing v3 EloChat & Writing15401744+10.3
ACCoRDGeneral QA68.178.2+10.1
Humanity's Last Exam DiamondGeneral QA19.129.1+10.0
CharXiv Extended (Mercor held-out)Multimodal68.978.9+10.0
Chess Puzzles (task version 2.0)Games48.058.0+10.0
Conceptual Reasoning IndexIndexes58.465.7+10.0
ArXivMath 06/2026Math57.667.4+9.7
LLM Debate BenchmarkChat & Writing15821514−9.7
AA-PersonalityBench Imagination indexChat & Writing31.541.0+9.6
Finance Agent v2 - Financial ModelingFinance & Accounting23.132.6+9.6
BrokenArXiv 05/2026Math28.519.0−9.5
ArXivMath 09/2026 (Native Harnesses)Math40.650.0+9.4
Humanity's Last ExamGeneral QA36.846.2+9.4
TxBench Antibody Discovery V1 (100 evaluations)Science39.749.0+9.3
ORCA-bench Hard RCA (Fireworks 324-task subset)Agents20.329.5+9.3
Code MigrationCode45.636.5−9.1
DistressBenchSafety79.870.7−9.0
Finance Agent v2 - All PassFinance & Accounting41.149.7+8.6
Finance Agent v2 - Earnings AnalysisFinance & Accounting72.981.4+8.5
HLE Extended (Mercor held-out)General QA45.336.8−8.5
SpatialBench V0 VerifiedScience75.467.0−8.4
GDPval-AA v2Professional Work16321464−8.3
AutomationBench-AA 1.0.6Agents68.960.9−8.0
TxBench Preclinical Pharmacology V1 Clean RerunsScience46.754.7+8.0
Finance Agent v2 - PrecedentsFinance & Accounting28.236.1+7.9
Extended NYT Connections 940 (Quadratic v1)Games89.697.4+7.8
Creative Writing v3 Rubric ScoreChat & Writing14.115.6+7.8
Vision Arena OverallMultimodal12631293+7.7
SAFE-Teen 106 worst branch: Urgent Safety (under-18 prompt)Safety27.234.8−7.6
SpeciEval (Refusal-Aware Retries, Historical)Safety97.489.9−7.5
SpeciEval (Pooled English Runs)Safety97.489.9−7.5
BioSecBench-Surveillance V1 Clean RerunsScience51.544.1−7.4
LLM Creative Story WritingChat & Writing-0.50.1+7.2
Finance Agent v2 - Disclosure AnalysisFinance & Accounting61.368.5+7.2
Terminal-Bench 4.0 (Artificial Analysis)Code26.819.7−7.1
SAFE-Teen 106: Urgent Safety (under-18 prompt)Safety7.914.9−7.0
GDPval-AA v2.1Professional Work16001442−7.0
MedCodeHealth & Medicine41.248.1+7.0
LMArena Agent Arena Chat Bash ReliabilityAgents-1.012.5+6.8
Terminal-Bench 2.1Code74.581.3+6.7
Quant-Arena (iterative total score)Finance & Accounting36.543.0+6.6
Financial Coaching BenchFinance & Accounting72.866.3−6.6
NeuroscienceBench V1 (91 evaluations)Science22.015.8−6.2
Vals Science IndexIndexes29.635.8+6.1
Vibe Code Bench v1.1Code84.778.7−6.1
HealthBench ProfessionalHealth & Medicine52.959.0+6.1
ProofBench v1.1Math54.048.0−6.0
Terminal-Bench 2.1 Extended (Mercor held-out)Code38.444.4+6.0
Harvey LAB: In-House Contracting (held-out)Legal0.36.3+6.0
Finance Agent v2 - ComparablesFinance & Accounting46.152.0+5.9
LMCAGeneral QA47.052.9+5.9
SAFE-Teen 106 worst branch: Relationships (under-18 prompt)Safety19.814.0+5.9
DTBenchGeneral QA89.995.7+5.8
Access to Justice BenchmarkLegal72.967.1−5.8
ITBench-AAAgents46.952.5+5.6
SWE-bench Multimodal (Vals v1, 480 Tasks)Code46.540.8−5.6
Finance Agent v2 - AdjustmentsFinance & Accounting49.254.8+5.6
BioMysteryBench (Vals Terminus 2)Science67.862.2−5.6
LMArena Agent Arena Chat Praise vs ComplaintAgents3.113.9+5.4
FrontierCode 1.1 Extended - ScoreCode56.651.2−5.4
CharXiv (descriptive and reasoning combined)Multimodal90.195.4+5.3
LMArena Agent Arena Code Bash ReliabilityAgents-0.210.2+5.2
FrontierCode 1.1 Extended - Pass RateCode62.056.9−5.2
ScBench-Long V1 Clean RerunsScience14.819.9+5.1
Excel Modeling Benchmark v1.1Finance & Accounting67.872.9+5.1
LMArena Agent Arena Work Bash ReliabilityAgents0.110.2+5.0
LMArena Code Arena WebDevCode16211583−5.0
Ramp Accounting BenchFinance & Accounting40.845.7+4.9
BioSecBench-Function V1 Clean RerunsScience40.836.0−4.9
FrontierMath Tier 4 (v2) - PrivateMath26.822.0−4.9
SciCode 1.0.1 (Artificial Analysis)Code51.956.6+4.7
HealthBenchHealth & Medicine61.957.1−4.7
VoxelBench Text RatingGames18611798−4.5
GameBench 2Games68.464.0−4.4
Vals Index v2Indexes57.962.3+4.4
RuneBenchGames5.76.0+4.3
Tax Agent BenchFinance & Accounting62.566.8+4.3
Terminal-Bench Science 0.1 (Vals)Science4.38.6+4.3
LMArena Agent Arena Bash RecoveryAgents7.9-0.7−4.3
LMArena Agent Arena Code Confirmed SuccessAgents6.6-1.8−4.2
EpiBench V1 Clean RerunsScience25.521.4−4.1
ARC-AGI-1 Semi-PrivateGeneral QA94.598.5+4.0
Mystery Game PuzzlesGames43.047.0+4.0
CritPt (avg@5)Science14.318.3+4.0
LMArena Text Arena EloChat & Writing14751497+3.9
LMArena Agent Arena Code Net Improvement (Steering Burden)Agents2.71.1−3.8
LMArena Agent Arena SteerabilityAgents-3.34.3+3.8
SpatialBench V1 Clean RerunsScience63.259.4−3.8
LegalBenchLegal83.387.0+3.7
BrowseComp (Mercor 130-question subset)Search79.483.1+3.7
Arena Image-to-WebDevCode16101594−3.6
APEX-AccountingFinance & Accounting48.151.7+3.6
Bedside BenchHealth & Medicine82.986.5+3.6
ArXivMath 08/2026 (Native Harnesses)Math43.940.4−3.5
Bug Hunt BenchCode20.617.1−3.5
GDP.pdfProfessional Work19.823.2+3.4
Vals Finance IndexIndexes57.160.5+3.4
Harvey's Legal Agent BenchmarkLegal6.710.0+3.3
APEX-1 (Mercor, mean score)Professional Work65.568.8+3.3
GDPval (Mercor 213-task rubric score)Professional Work86.783.5−3.2
Concept Synth INDUCTION Challenge64 Round 1General QA17.214.1−3.1
LMArena Agent Arena Chat Confirmed SuccessAgents4.911.0+3.0
SciCode (main problems)Science11.214.2+3.0
LMArena Agent Arena Work Confirmed SuccessAgents5.7-0.3−3.0
LMArena Agent Arena Chat Steering BurdenAgents-0.55.4+2.9
RareDiseaseBench V1 (35 evaluations)Science0.02.9+2.9
SAGE v1.1Math85.488.2+2.9
Design Arena Overall Code CategoriesCode13161299−2.8
Longform Creative WritingChat & Writing74.076.8+2.8
Furniture Assembly BenchmarkMultimodal34.231.7−2.5
MMLU ProGeneral QA88.290.7+2.5
Legal Research BenchLegal41.338.9−2.4
SpatialBench-Long V1 Clean RerunsScience26.228.6+2.4
SWE-bench VerifiedCode82.980.6−2.3
Epiq LAER CorporateBenchProfessional Work75.377.6+2.3
Immigration BenchLegal65.663.4−2.2
Political Science LLM BenchmarkProfessional Work66.568.7+2.2
Revelo Code IndexCode38.936.7−2.2
BioSecBench-RespiratoryVirus-Surveillance V1 (110 evaluations)Science5.27.3+2.1
LMArena Agent Arena Code Steering BurdenAgents2.3-1.8−2.0
Concept Synth INDUCTION Challenge100 Round 1General QA11.09.0−2.0
TaxCalcBench TY25Finance & Accounting4.06.0+2.0
AA-PersonalityBench Informativeness indexChat & Writing79.181.0+1.9
Epoch Capabilities Index (ECI)Indexes156159+1.8
Harvey LAB-AA v1.1 (hallucination-gated)Legal1.70.0−1.7
Artificial Analysis Intelligence Index v4.3Indexes39.541.2+1.6
Decision-Bench v1.1Professional Work25.326.9+1.6
AIME 2025Math98.396.7−1.6
Artificial Analysis Intelligence Index v4.3.2Indexes39.540.9+1.5
WeirdML v3Science5.97.4+1.4
Terminal-Bench-Science 0.1Science9.010.5+1.4
LMArena Agent Arena Work Net Improvement (Steering Burden)Agents2.52.0−1.4
Westworld Finance Diligence BenchFinance & Accounting48.349.7+1.4
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine14.713.3−1.3
ComplexConstraintsChat & Writing49.748.4−1.3
Medical Long Context Reasoning (MLCR-AA)Health & Medicine22.821.7−1.1
FrontierMath Tiers 1-3 (v2) - PrivateMath67.468.4+1.1
Arena Alignment IndexSafety74.875.8+1.0
Public Benefits Bench v1.1Professional Work64.365.3+1.0
MedScribeHealth & Medicine85.584.5−1.0
DualEntry Accounting AI BenchmarkFinance & Accounting84.283.2−1.0
MineBench — Bradley-Terry RatingGames18221806−1.0
LMArena Agent Arena Work Steering BurdenAgents1.50.1−0.7
TissueMicroscopyBench V1 (49 evaluations)Science2.73.4+0.7
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting67.468.1+0.7
TxBench-Antibody-Development V1 (60 evaluations)Science14.415.0+0.6
OTIS Mock AIME 2024-2025Math98.398.9+0.6
ProgramBenchCode0.51.0+0.5
SAFE-Teen 106: Relationships (under-18 prompt)Safety4.23.7+0.5
DeepSWE v1.1Code74.273.8−0.4
LMArena Agent Arena Net ImprovementAgents3.93.8−0.3
Finance Agent v2.1Finance & Accounting65.465.8+0.3
VariantBench V1 Clean RerunsScience31.131.4+0.3
TxBench Oligo Discovery V1 (113 evaluations)Science34.835.1+0.3
xbench-ScienceQAScience72.672.4−0.2
LMArena Agent Arena Work Praise vs ComplaintAgents3.73.3−0.2
LMArena Agent Arena Code Praise vs ComplaintAgents3.02.7−0.1
LMArena Agent Arena Chat Tool HallucinationAgents0.30.2−0.1
LMArena Agent Arena Code Tool HallucinationAgents0.50.60.0
Artificial Analysis Progress Index v1Indexes3203200.0
LMArena Agent Arena Work Tool HallucinationAgents0.50.50.0
ArXivMath 05/2026Math58.358.30.0
GeneBench-Pro Public Release SubsetScience11.011.00.0
Terminal-Bench 4.0Code19.719.70.0

DeepSeek V4.1 FlashvsGemini 3.8 Flashnormalised score marginsData Nov 15, 2026Data: Vals AI, Artificial Analysis, Arena and 52 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…