Atlas

Compare Models

Opus 4.8 vs DeepSeek V4.1 Flash 1017

DeepSeek V4.1 Flash 1017 leads by 1.2 capability points (≈95% interval −1.0 to +3.4): statistically tied · P(higher) ≈ 86%

Wins 53 of 92 shared benchmarks (39 lost)

Capability vs. relative cost
Opus 4.8DeepSeek V4.1 Flash 1017Closed 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.

Opus 4.8vsDeepSeek V4.1 Flash 1017capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins39 – 53 · 92 shared

Comparing matching setups only (7 benchmarks excluded) · suites counted once
← Opus 4.8DeepSeek V4.1 Flash 1017 →

Opus 4.8vsDeepSeek V4.1 Flash 1017wins by categoryData Nov 15, 2026Data: Vals AI, SecureBio, Artificial Analysis and 30 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is DeepSeek V4.1 Flash 1017 minus Opus 4.8.

HardGeoBench Avg DistanceMultimodal12453201−60.4
LisanBench Difficulty-WeightedGames26947733+54.0
HardGeoBench Country AccuracyMultimodal53.815.0−38.7
LisanBench Path LengthGames946321107+36.2
HardGeoBench <=100miMultimodal36.37.5−28.7
Concept Synth INDUCTION Challenge64 Round 1General QA6.332.8+26.6
HardGeoBench <=10miMultimodal27.53.8−23.8
DeepsecBench — Revalidated V1 (233 issues)Code6.930.0+23.1
Bug Hunt BenchCode14.337.3+23.0
AutomationBench-AA 1.0.6Agents45.668.5+22.9
DeepSWE v1.1Code54.477.0+22.6
WCB (SecureBio)Science66.749.6−17.1
ZeroBench (Main questions, pass@5, with tools)Multimodal34.051.0+17.0
APEX-Agents 1.1Professional Work64.479.4+15.0
LMArena Code Arena WebDevCode15551664+14.2
Mystery Game PuzzlesGames36.049.0+13.0
Terminal-Bench 4.0 (Artificial Analysis)Code21.733.8+12.1
AA-Omniscience IndexIndexes28.84.7−12.0
BabyVision with PythonMultimodal81.292.5+11.3
NoCoT-Bench NCRI 15.2General QA10281.6−11.2
NoCoT-Bench NCKI (kspine v3)General QA117102−10.8
Bio Capabilities Index (SecureBio)Indexes164149−10.7
SimpleQA Verified (Epoch anti-abstention)General QA53.042.5−10.5
Tax Agent BenchFinance & Accounting64.974.9+10.0
Finance Agent v2 - AdjustmentsFinance & Accounting48.458.4+10.0
FrontierMath Tier 4 (v2) - PrivateMath56.146.3−9.8
FrontierMath Tiers 1-3 (v2) - PrivateMath80.071.2−8.8
BuseyBench SVGMultimodal5.24.3−8.6
Terminal-Bench 4.0Code23.231.8+8.6
Chartography (with tools)Multimodal75.083.0+8.0
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting66.474.1+7.7
Excel Modeling BenchmarkFinance & Accounting69.461.8−7.6
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting71.679.2+7.5
GameBench 2Games62.569.9+7.4
VCT v2 (SecureBio)Science45.938.5−7.4
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine19.426.7+7.2
Humanity's Last ExamGeneral QA49.842.6−7.2
GPQA-Bio (SecureBio)Science80.987.8+6.9
Humanity's Sixth SenseMultimodal23.530.1+6.6
EnterpriseBench: CoreCraftProfessional Work52.358.5+6.2
Design Arena Overall Code CategoriesCode12521289+6.1
Chess Puzzles (task version 2.0)Games52.046.0−6.0
ZeroBench (Main questions, pass@5)Multimodal17.023.0+6.0
ARC-AGI-2General QA72.178.1+6.0
FrontierCode 2.0 - Pass RateCode17.123.0+5.9
GDP.pdf (Artificial Analysis, All-pass)Professional Work22.817.2−5.6
APEX-1 (Mercor, mean score)Professional Work62.768.3+5.6
VCT (SecureBio)Science48.242.9−5.3
AA-Briefcase v1.1 EloAgents13201416+5.2
RedlineBenchLegal39.744.9+5.2
APEX-SWECode43.949.1+5.2
Wiz Cyber Model Arena v3.9Cybersecurity65.460.3−5.1
Vibe Code Bench v1.1Code82.787.7+5.0
LAB-Bench ProtocolQA (SecureBio)Science76.981.9+5.0
FrontierCode 2.0 - ScoreCode15.320.2+4.9
Finance Agent v2 - Market AnalysisFinance & Accounting67.262.7−4.5
WMDP-Bio (SecureBio)Science92.387.9−4.4
ARC-AGI-3 (Semi-Private)General QA1.55.8+4.3
PubMedQA (SecureBio)Health & Medicine75.980.0+4.1
CharXiv ReasoningMultimodal80.577.0−3.5
Artificial Analysis Intelligence Index v4.3.2Indexes41.845.1+3.3
Finance Agent v2 - Earnings AnalysisFinance & Accounting72.175.3+3.2
ALE-BenchCode21022012−3.2
Code MigrationCode47.344.2−3.0
Finance Agent v2 - ComparablesFinance & Accounting44.947.5+2.7
HPCT (SecureBio)Science63.160.5−2.6
MBCT (SecureBio)Science56.453.9−2.5
LiveBench 2.0Agents76.879.2+2.4
Humanity's Last Exam DiamondGeneral QA23.120.8−2.3
ReactBench (39 tasks)Code32.334.4+2.1
ARC-AGI-1 Semi-PrivateGeneral QA92.594.5+2.0
Artificial Analysis Progress Index v1Indexes318324+1.9
Finance Agent v2 - All PassFinance & Accounting43.545.2+1.7
SciCode 1.0.1 (Artificial Analysis)Code54.455.9+1.5
NL2Repo - release-reported repository generation scoreCode69.768.2−1.5
Vals Index v2.1Indexes55.156.6+1.5
Legal Research BenchLegal43.845.2+1.4
HealthBench Professional (Length-Adjusted)Health & Medicine53.051.7−1.3
OTIS Mock AIME 2024-2025Math98.399.4+1.1
LAB-Bench CloningScenarios (SecureBio)Science95.994.8−1.1
BrowseCompSearch84.383.4−0.9
Finance Agent v2 - PrecedentsFinance & Accounting25.626.5+0.8
Ruby LLM Benchmarks — Program FixerCode72.672.0−0.7
Finance Agent v2 - Financial ModelingFinance & Accounting23.322.8−0.6
Finance Agent v2 - Disclosure AnalysisFinance & Accounting65.766.3+0.6
PerceptionBenchMultimodal47.247.7+0.5
Epoch Capabilities Index (ECI)Indexes160159−0.5
Harvey's Legal Agent BenchmarkLegal9.69.2−0.4
LAB-Bench LitQA2 (SecureBio)Science48.749.0+0.3
GBENCH One-shot Coding GScoreGames54.053.8−0.2
MMLU-Bio (SecureBio)Science90.990.7−0.2
MMMU-Pro with PythonMultimodal82.782.5−0.2

Opus 4.8vsDeepSeek V4.1 Flash 1017normalised score marginsData Nov 15, 2026Data: Vals AI, SecureBio, Artificial Analysis and 30 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…