Atlas

Compare Models

DeepSeek V4.1 Flash 1017 vs GPT-5.5

DeepSeek V4.1 Flash 1017 leads by 0.9 capability points (≈95% interval −1.3 to +3.2): statistically tied · P(higher) ≈ 79%

GPT-5.5 wins 43 of 78 shared benchmarks (35 lost)

Capability vs. relative cost
DeepSeek V4.1 Flash 1017GPT-5.5Closed 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 1017vsGPT-5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins35 – 43 · 78 shared

Comparing matching setups only (24 benchmarks excluded) · suites counted once
← DeepSeek V4.1 Flash 1017GPT-5.5 →

DeepSeek V4.1 Flash 1017vsGPT-5.5wins by categoryData Nov 15, 2026Data: Vals AI, Artificial Analysis, SecureBio and 25 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-5.5 minus DeepSeek V4.1 Flash 1017.

LisanBench Difficulty-WeightedGames77333316−47.3
LisanBench Path LengthGames211079845−35.0
FrontierMath Tier 4 (v2) - PrivateMath46.372.5+26.2
AutomationBench-AA 1.0.6Agents68.547.3−21.2
SimpleQA Verified (Epoch anti-abstention)General QA42.563.0+20.5
Terminal-Bench 4.0 (Artificial Analysis)Code28.89.1−19.7
AA-Briefcase v1.1 EloAgents14161087−18.0
Vibe Code Bench v1.1Code87.769.8−17.9
Chess Puzzles (task version 2.0)Games46.063.0+17.0
BuseyBench SVGMultimodal4.36.0+16.5
Tax Agent BenchFinance & Accounting74.960.5−14.5
FrontierMath Tiers 1-3 (v2) - PrivateMath71.285.3+14.0
Finance Agent v2 - AdjustmentsFinance & Accounting58.445.7−12.6
DeepSWE v1.1Code77.064.4−12.6
APEX-SWECode49.136.6−12.5
DAXBench v2Code77.788.3+10.6
LiveBench 2.0Agents79.268.7−10.5
FoodTruck BenchGames5328961408+10.4
NoCoT-Bench NCRI 15.2General QA81.6100+10.2
WCB (SecureBio)Science49.659.7+10.1
ZeroBench (Main questions, pass@5, with tools)Multimodal51.041.0−10.0
GameBench 2Games69.959.9−10.0
VCT (SecureBio)Science42.952.3+9.4
APEX-Agents 1.1Professional Work79.470.3−9.1
DeepsecBench — Revalidated V1 (233 issues)Code30.021.0−8.9
BabyVision with PythonMultimodal92.583.6−8.9
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting74.165.6−8.5
GBENCH One-shot Coding GScoreGames53.862.2+8.4
PerceptionBenchMultimodal47.755.8+8.1
AA-Omniscience IndexIndexes4.720.5+7.9
Finance Agent v2 - Disclosure AnalysisFinance & Accounting66.358.9−7.4
EnterpriseBench: CoreCraftProfessional Work58.551.3−7.2
Humanity's Last Exam DiamondGeneral QA20.827.9+7.1
CharXiv ReasoningMultimodal77.084.1+7.1
Mystery Game PuzzlesGames49.056.0+7.0
ARC-AGI-2General QA78.185.0+6.9
VCT v2 (SecureBio)Science38.545.3+6.8
Artificial Analysis Intelligence Index v4.3.2Indexes45.138.4−6.7
Bio Capabilities Index (SecureBio)Indexes149158+6.4
NoCoT-Bench NCKI (kspine v3)General QA102111+6.3
HPCT (SecureBio)Science60.566.8+6.3
Finance Agent v2 - All PassFinance & Accounting45.239.6−5.7
Ruby LLM Benchmarks — Program FixerCode72.077.6+5.6
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting79.273.7−5.5
Harvey's Legal Agent BenchmarkLegal9.23.8−5.4
GDP.pdf (Artificial Analysis, All-pass)Professional Work17.222.6+5.4
ZeroBench Main Mean pass@1 (five samples)Multimodal11.817.2+5.4
ARC-AGI-3 (Semi-Private)General QA5.80.4−5.4
Design Arena Overall Code CategoriesCode12891257−5.3
Legal Research BenchLegal45.240.4−4.8
Finance Agent v2 - PrecedentsFinance & Accounting26.521.9−4.6
LAB-Bench ProtocolQA (SecureBio)Science81.977.9−4.0
Finance Agent v2 - ComparablesFinance & Accounting47.543.7−3.9
ALE-BenchCode20122108+3.4
MBCT (SecureBio)Science53.957.1+3.2
APEX-Accounting 1.1Finance & Accounting59.356.3−3.0
Excel Modeling BenchmarkFinance & Accounting61.864.5+2.7
LAB-Bench CloningScenarios (SecureBio)Science94.897.3+2.5
Finance Agent v2 - Market AnalysisFinance & Accounting62.760.3−2.4
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine26.728.9+2.2
Finance Agent v2 - Earnings AnalysisFinance & Accounting75.373.2−2.1
AA-LCR v1.1Search82.784.3+1.7
MMMU ProMultimodal79.781.2+1.4
CritPt (avg@5)Science24.025.4+1.4
Artificial Analysis Progress Index v1Indexes324320−1.3
Humanity's Last ExamGeneral QA42.643.8+1.2
BrowseCompSearch83.484.4+1.0
Code MigrationCode44.245.2+1.0
RedlineBenchLegal44.944.1−0.8
LAB-Bench LitQA2 (SecureBio)Science49.049.8+0.8
Epoch Capabilities Index (ECI)Indexes159160+0.6
OTIS Mock AIME 2024-2025Math99.4100.0+0.6
ARC-AGI-1 Semi-PrivateGeneral QA94.595.0+0.5
PubMedQA (SecureBio)Health & Medicine80.080.5+0.5
GPQA-Bio (SecureBio)Science87.887.4−0.4
SciCode 1.0.1 (Artificial Analysis)Code55.956.1+0.2
Finance Agent v2 - Financial ModelingFinance & Accounting22.822.9+0.2
APEX-1 (Mercor, mean score)Professional Work68.368.4+0.1

DeepSeek V4.1 Flash 1017vsGPT-5.5normalised score marginsData Nov 15, 2026Data: Vals AI, Artificial Analysis, SecureBio and 25 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…