Atlas

Compare Models

DeepSeek V4.1 Flash 1017 vs Gemini 3.8 Flash

DeepSeek V4.1 Flash 1017 leads by 0.3 capability points (≈95% interval −2.1 to +2.6): statistically tied · P(higher) ≈ 58%

Wins 68 of 133 shared benchmarks (64 lost, 1 tied)

Capability vs. relative cost
DeepSeek V4.1 Flash 1017Gemini 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 Flash 1017vsGemini 3.8 Flashcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins68 – 64 · 133 shared, 1 tied

Comparing matching setups only (17 benchmarks excluded) · suites counted once
← DeepSeek V4.1 Flash 1017Gemini 3.8 Flash →

DeepSeek V4.1 Flash 1017vsGemini 3.8 Flashwins by categoryData Nov 15, 2026Data: Artificial Analysis, Vals AI, SecureBio and 31 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 1017.

CyberGym-E2E-AA (Artificial Analysis)Cybersecurity76.38.4−67.9
AA-PersonalityBench Transparency indexChat & Writing86.723.8−63.5
AA-PersonalityBench Slop indexChat & Writing39.171.4+32.7
AA-PersonalityBench Warmth indexChat & Writing27.760.0+32.7
ARC-AGI-3 (Semi-Private)General QA5.835.0+29.2
Code Migration v1.1Code67.538.6−28.9
SimpleQA Verified (Epoch anti-abstention)General QA42.569.7+27.2
AA-PersonalityBench Expressiveness indexChat & Writing52.479.1+26.9
FrontierMath Tier 4 (v2) - PrivateMath46.322.0−24.4
NoCoT-Bench NCRI 15.2General QA81.6126+24.3
Bug Hunt BenchCode40.717.1−23.5
BrokenArXiv 09/2026 (Native Harnesses)Math49.726.7−23.0
BuseyBench SVGMultimodal4.36.6+22.8
VibeVideoBench (overall quality, no video = 0)Code3.92.8−22.3
BrokenArXiv 08/2026 (Native Harnesses)Math46.727.4−19.3
Concept Synth INDUCTION Challenge64 Round 1General QA32.814.1−18.7
RiemannBench (November 2026 rescoring)Math38.456.8+18.4
FrontierML ($200 wallet)Agents23.55.7−17.8
VoxelBench Text RatingGames20471798−17.7
CWE-Bench-AA (Artificial Analysis)Cybersecurity69.252.5−16.7
Finance Agent v2 - Market AnalysisFinance & Accounting62.778.5+15.8
AA-PersonalityBench Organization indexChat & Writing88.673.3−15.4
NoCoT-Bench NCKI (kspine v3)General QA102121+13.2
CritPt v2 (agentic, avg@5)Science52.365.1+12.9
APEX-SWECode49.136.4−12.7
AA-Omniscience IndexIndexes4.729.6+12.4
Terminal-Bench 4.0Code31.819.7−12.1
Concept Synth INDUCTION Challenge100 Round 1General QA21.09.0−12.0
Chess Puzzles (task version 2.0)Games46.058.0+12.0
LiveBench 2.0Agents79.267.3−11.9
AA-Briefcase v1.1 EloAgents14161204−11.6
Humanity's Sixth SenseMultimodal30.141.6+11.5
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine26.715.6−11.1
ARC-AGI-2General QA78.189.2+11.1
LMArena Code Arena WebDevCode16641583−10.6
DeepsecBench-AA (Artificial Analysis)Cybersecurity24.914.5−10.4
Excel Modeling BenchmarkFinance & Accounting61.872.2+10.4
Finance Agent v2 - Financial ModelingFinance & Accounting22.832.6+9.9
Finance Agent v2 - PrecedentsFinance & Accounting26.536.1+9.7
AA-PersonalityBench Conviction indexChat & Writing94.384.8−9.6
AA-PersonalityBench Informativeness indexChat & Writing71.481.0+9.6
DAXBench v2Code77.787.2+9.6
Terminal-Bench 4.0 (Artificial Analysis)Code28.819.7−9.1
Vibe Code Bench v1.1Code87.778.7−9.1
Humanity's Last Exam DiamondGeneral QA20.829.1+8.3
Tax Agent BenchFinance & Accounting74.966.8−8.1
DAYJOB: Finance (October 2026 rescoring)Finance & Accounting12.54.5−8.0
Code MigrationCode44.236.5−7.7
AutomationBench-AA 1.0.6Agents68.560.9−7.6
WCB (SecureBio)Science49.657.1+7.5
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal18.811.4−7.4
τ³-Banking (Mercor run, bm25_grep retrieval)Agents41.648.8+7.2
VCT v2 (SecureBio)Science38.545.2+6.7
FrontierCode 1.1 Main (Artificial Analysis)Code43.737.1−6.5
APEX-Agents 1.1Professional Work79.472.9−6.5
Legal Research BenchLegal45.238.9−6.2
Finance Agent v2 - Earnings AnalysisFinance & Accounting75.381.4+6.1
Access to Justice BenchmarkLegal73.167.1−6.0
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting74.168.1−6.0
ALE-BenchCode20122182+6.0
GameBench 2Games69.964.0−5.9
CritPt (avg@5)Science24.018.3−5.7
GDP.pdf (Artificial Analysis, All-pass)Professional Work17.222.8+5.6
Vals Index v2.2Indexes61.356.0−5.3
LAB-Bench CloningScenarios (SecureBio)Science94.8100.0+5.2
MBCT (SecureBio)Science53.948.9−5.0
GBENCH One-shot Coding GScoreGames53.858.6+4.9
MMMU ProMultimodal79.784.5+4.8
ArXivMath 09/2026 (Native Harnesses)Math45.350.0+4.7
τ³-Banking (Fireworks, banking_knowledge)Agents46.851.4+4.6
Finance Agent v2 - ComparablesFinance & Accounting47.552.0+4.5
Finance Agent v2 - All PassFinance & Accounting45.249.7+4.5
Visa Agentic Commerce BenchmarkAgents57.853.3−4.4
Artificial Analysis Intelligence Index v4.3.2Indexes45.140.9−4.2
HoneyBench v0.2 (hack rate)Safety76.480.5−4.1
ARC-AGI-1 Semi-PrivateGeneral QA94.598.5+4.0
DAYJOB: Healthcare (October 2026 rescoring)Health & Medicine6.02.0−4.0
VCT (SecureBio)Science42.946.9+4.0
AA-PersonalityBench Imagination indexChat & Writing37.241.0+3.8
LMArena Agent Arena Bash ReliabilityAgents2.49.9+3.8
Excel Modeling Benchmark v1.1Finance & Accounting69.272.9+3.7
RedlineBenchLegal44.941.3−3.6
Humanity's Last ExamGeneral QA42.646.2+3.6
FrontierPhysics v0.1Science3.60.0−3.6
Finance Agent v2 - AdjustmentsFinance & Accounting58.454.8−3.5
ArXivMath 08/2026 (Native Harnesses)Math43.940.4−3.5
Arena Image-to-WebDevCode16091594−3.4
Bio Capabilities Index (SecureBio)Indexes149153+3.3
MKA (Artificial Analysis)Professional Work38.535.3−3.3
DeepSWE v1.1Code77.073.8−3.2
Finance Agent v2.1Finance & Accounting62.965.8+2.8
FrontierMath Tiers 1-3 (v2) - PrivateMath71.268.4−2.8
LMArena Agent Arena Steering BurdenAgents4.0-1.3−2.7
HANDBOOK.md Agents v1.1Professional Work24.026.5+2.5
LAB-Bench LitQA2 (SecureBio)Science49.051.5+2.5
VIALS-AA (pass@3)Multimodal66.564.0−2.5
Political Science LLM BenchmarkProfessional Work66.268.7+2.5
Epiq LAER CorporateBenchProfessional Work75.277.6+2.4
Finance Agent v2 - Disclosure AnalysisFinance & Accounting66.368.5+2.2
LAB-Bench ProtocolQA (SecureBio)Science81.984.0+2.1
HLAB-Vals-1.1Legal14.212.1−2.1
Vals Legal IndexIndexes71.969.8−2.1
Artificial Analysis Intelligence Index v5.0Indexes28.530.5+2.0
Mystery Game PuzzlesGames49.047.0−2.0
HPCT (SecureBio)Science60.562.4+1.9
FrontierCode 2.0 (Artificial Analysis)Code16.818.7+1.9
LMArena Agent Arena Praise vs ComplaintAgents1.24.9+1.9
Artificial Analysis Intelligence Index v5.0.1Indexes32.630.8−1.7
Vals Index v2.1Indexes56.654.9−1.7
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting79.280.9+1.7
Design Arena Overall Code CategoriesCode12891299+1.7
DAYJOB: AccountingFinance & Accounting7.66.0−1.6
WMDP-Bio (SecureBio)Science87.989.4+1.5
AA-LCR v1.1Search82.784.0+1.3
VIALS-AA (average of 3)Multimodal53.252.0−1.2
Artificial Analysis Progress Index v1Indexes324320−1.2
DualEntry Accounting AI BenchmarkFinance & Accounting82.283.2+1.0
Bedside BenchHealth & Medicine85.786.5+0.8
Harvey's Legal Agent BenchmarkLegal9.210.0+0.8
SciCode 1.0.1 (Artificial Analysis)Code55.956.6+0.7
VIALS-AA (pass^3)Multimodal40.441.0+0.6
LMArena Agent Arena Net Improvement (Steering Burden)Agents2.02.3+0.6
OTIS Mock AIME 2024-2025Math99.498.9−0.6
Epoch Capabilities Index (ECI)Indexes159159−0.5
LMArena Agent Arena Confirmed SuccessAgents2.31.3−0.5
APEX-1 (Mercor, mean score)Professional Work68.368.8+0.5
APEX-Accounting 1.1Finance & Accounting59.358.8−0.5
Terminal-Bench-Science 0.1Science10.510.0−0.5
Vals Finance IndexIndexes60.960.5−0.4
Build-A-Chip-Bench (20-task index run)Agents0.50.1−0.4
DAYJOB: TaxFinance & Accounting1.20.8−0.4
LMArena Agent Arena Tool HallucinationAgents0.30.4+0.1
EnterpriseBench: CoreCraftProfessional Work58.558.50.0

DeepSeek V4.1 Flash 1017vsGemini 3.8 Flashnormalised score marginsData Nov 15, 2026Data: Artificial Analysis, Vals AI, SecureBio and 31 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…