Atlas

Compare Models

Opus 4.8 vs DeepSeek V4.1 Flash

Opus 4.8 leads by 0.3 capability points (≈95% interval −1.6 to +2.2): statistically tied · P(higher) ≈ 64%

Wins 92 of 160 shared benchmarks (65 lost, 3 tied)

Capability vs. relative cost
Opus 4.8DeepSeek V4.1 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.

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

Benchmark wins92 – 65 · 160 shared, 3 tied

Comparing matching setups only (35 benchmarks excluded) · suites counted once
← Opus 4.8DeepSeek V4.1 Flash →

Opus 4.8vsDeepSeek V4.1 Flashwins by categoryData Nov 15, 2026Data: Vals AI, Mercor, benchmarks.bio and 42 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 minus Opus 4.8.

Surface Evolver BenchAgents87.546.3−41.2
KernelBench Hard Normalized (RTX PRO 6000)Code32.669.7+37.1
BullshitBench v2 LeaderboardSafety94.057.0−37.0
FrontierMath Tier 4 (v2) - PrivateMath56.126.8−29.3
LLM Chess EloGames4501087+26.1
A-Fantasia (Valid Attempts with Retries)Games34.759.3−24.7
AutomationBench-AA 1.0.6Agents45.668.9+23.3
RiemannBenchMath47.224.0−23.2
Medical Long Context Reasoning (MLCR-AA)Health & Medicine45.622.8−22.8
KernelBench CUDA Normalized (RTX PRO 6000)Code24.947.0+22.1
DeepSWE v1.1Code54.474.2+19.8
SWE-bench Verified Extended (Mercor held-out)Code63.382.7+19.4
SkillsBench without SkillsAgents43.761.7+18.0
ArXivMath 04/2026Math62.545.0−17.5
AA-Omniscience IndexIndexes28.8-5.3−17.0
TaxCalcBench TY25Finance & Accounting20.04.0−16.0
Apex (MathArena)Math81.365.6−15.7
Wiz Cyber Model Arena v3.9Cybersecurity65.450.2−15.2
ZeroBench (Main questions, pass@5, with tools)Multimodal34.049.0+15.0
BabyVisionMultimodal28.443.3+14.9
ComplexConstraintsChat & Writing35.649.7+14.1
LLM Creative Story WritingChat & Writing0.8-0.5−14.1
APEX-Agents 1.1Professional Work64.450.5−13.9
Chartography (with tools)Multimodal65.078.9+13.9
Conceptual Reasoning IndexIndexes68.358.4−13.6
SpeciEval (Pooled English Runs)Safety84.397.4+13.1
SpeciEval (Refusal-Aware Retries, Historical)Safety84.397.4+13.0
Humanity's Last ExamGeneral QA49.836.8−13.0
SWE Atlas Codebase QnA Extended (Mercor held-out)Code24.036.7+12.7
FrontierMath Tiers 1-3 (v2) - PrivateMath80.067.4−12.6
ArXivMath 06/2026Math70.057.6−12.3
Excel Modeling BenchmarkFinance & Accounting69.457.2−12.2
MedCodeHealth & Medicine53.241.2−12.0
VariantBench V1 Clean RerunsScience42.431.1−11.3
BuseyBench SVGMultimodal5.24.1−11.2
TxBench Preclinical Pharmacology V1 Clean RerunsScience57.746.7−11.0
Concept Synth INDUCTION Challenge64 Round 1General QA6.317.2+10.9
EnterpriseBench: CoreCraftProfessional Work52.341.5−10.8
Design Arena Overall Code CategoriesCode12521316+10.6
SkillsBench with SkillsAgents59.269.8+10.6
LMCAGeneral QA57.547.0−10.5
MetagenomicsBench V1Science48.338.0−10.3
AdvancedIFChat & Writing73.784.0+10.3
MetagenomicsBench V0Science48.338.0−10.3
SimpleQA Verified (Epoch anti-abstention)General QA53.042.9−10.1
AA-AnalystAgent (pass^5)Professional Work45.035.0−10.0
LMArena Agent Arena Net ImprovementAgents6.93.9−9.9
ALE-BenchCode21021829−9.7
ACCoRDGeneral QA77.468.1−9.3
SWE Atlas - Codebase QnACode44.453.5+9.1
MedXpertQA MM (200-question set)Health & Medicine72.563.5−9.0
SpatialBench V0 VerifiedScience66.475.4+9.0
AutoformBenchMath44.052.8+8.8
BlenderWorldCode57.065.7+8.7
DeepResearch Bench II (Mercor 130-task set)Search44.653.2+8.6
LMArena Code Arena WebDevCode15551621+8.5
DistressBenchSafety71.479.8+8.3
Furniture Assembly BenchmarkMultimodal42.534.2−8.3
APEX-SWECode43.951.9+8.0
CadQueryEval (Corrected Geometry)Code68.076.0+8.0
Senior SWE-BenchCode30.522.7−7.8
CharXiv Extended (Mercor held-out)Multimodal76.768.9−7.8
KernelBench Mega Normalized (RTX PRO 6000)Code40.648.2+7.6
AA-LCR Extended (Mercor held-out)Search63.270.8+7.6
Terminal-Bench 2.1Code82.790.3+7.6
Terminal-Bench 4.0Code23.631.2+7.6
HealthBench ProfessionalHealth & Medicine60.352.9−7.4
SpatialBench-Long V1 Clean RerunsScience33.326.2−7.1
GeneBench-Pro Public Release SubsetScience18.011.0−7.0
Mystery Game PuzzlesGames36.043.0+7.0
LLM Debate BenchmarkChat & Writing16311582−6.9
SAGEMath54.847.9−6.9
ArXivMath 05/2026Math65.058.3−6.7
TxBench Antibody Discovery V0Science45.639.0−6.6
LMArena Agent Arena SteerabilityAgents9.6-3.3−6.4
MMMU-Pro Extended (Mercor held-out)Multimodal31.425.0−6.4
Bug Hunt BenchCode14.320.6+6.3
TxBench Antibody Discovery V1 (100 evaluations)Science33.339.7+6.3
AA-Briefcase EloAgents13161424+6.3
ReactBench (39 tasks): Fixing ReactCode12.518.8+6.3
GDPval-AA v2.1Professional Work14611600+6.1
SpatialBench-Long V0 VerifiedScience28.834.9+6.1
GBENCH One-shot Coding GScoreGames54.048.0−6.1
GameBench 2Games62.568.4+6.0
Vision Arena OverallMultimodal12861263−5.9
BioSecBench-Function V0Science44.638.7−5.9
BrowseComp (Mercor 130-question subset)Search73.779.4+5.7
SciCode (main problems)Science16.911.2−5.7
Terminal-Bench 2.1 Extended (Mercor held-out)Code32.738.4+5.7
BrokenArXiv 05/2026Math23.028.5+5.5
TxBench Oligo Discovery V0Science39.433.9−5.5
GDPval-AA v2Professional Work15291632+5.1
DTBenchGeneral QA94.989.9−5.1
Terminal-Bench 4.0 (Artificial Analysis)Code21.726.8+5.1
BioSecBench-Surveillance V1 Clean RerunsScience46.951.5+4.6
Finance Agent v2 - Disclosure AnalysisFinance & Accounting65.761.3−4.4
NL2Repo - release-reported repository generation scoreCode69.765.4−4.3
Terminal-Bench Science 0.1 (Vals)Science4.30.0−4.3
GDP.pdfProfessional Work24.019.8−4.2
HLE Extended (Mercor held-out)General QA49.445.3−4.1
Chess Puzzles (task version 2.0)Games52.048.0−4.0
Humanity's Last Exam DiamondGeneral QA23.119.1−4.0
Finance Agent v2 - Market AnalysisFinance & Accounting67.263.4−3.9
Public Benefits Bench v1.1Professional Work68.164.3−3.9
SUPERChemScience60.356.7−3.6
FrontierCode 1.1 Extended - Pass RateCode65.562.0−3.5
MedXpertQA MM Extended (Mercor held-out)Health & Medicine51.648.4−3.2
EpiBenchScience39.042.1+3.1
CharXiv (descriptive and reasoning combined)Multimodal93.290.1−3.1
HealthBenchHealth & Medicine58.861.9+3.1
Vals Index v2Indexes60.957.9−3.0
FrontierCode 1.1 Extended - ScoreCode59.656.6−3.0
Harvey's Legal Agent BenchmarkLegal9.66.7−2.9
APEX-1 (Mercor, mean score)Professional Work62.765.5+2.8
Epoch Capabilities Index (ECI)Indexes160156−2.8
TxBench Oligo Discovery V1 (113 evaluations)Science37.534.8−2.6
SpatialBench V1 Clean RerunsScience65.863.2−2.6
Finance Agent v2 - PrecedentsFinance & Accounting25.628.2+2.6
SciCode 1.0.1 (Artificial Analysis)Code54.451.9−2.5
Finance Agent v2 - All PassFinance & Accounting43.541.1−2.5
Tax Agent BenchFinance & Accounting64.962.5−2.4
Artificial Analysis Intelligence Index v4.3Indexes42.039.5−2.4
Legal Research BenchLegal43.841.3−2.4
ScBench V1 Clean RerunsScience57.154.7−2.4
Artificial Analysis Intelligence Index v4.3.2Indexes41.839.5−2.3
BrokenArXiv 06/2026Math41.739.4−2.3
BioSecBench-Function V1 Clean RerunsScience43.140.8−2.2
GDPval (Mercor 213-task rubric score)Professional Work84.586.7+2.2
MineBench — Bradley-Terry RatingGames17861822+2.2
Vibe Code Bench v1.1Code82.784.7+2.0
ARC-AGI-1 Semi-PrivateGeneral QA92.594.5+2.0
SimpleBenchGeneral QA64.866.7+1.9
YC-BenchGames17113891657236−1.9
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting71.669.8−1.8
MMLU ProGeneral QA90.088.2−1.8
ScBench-Long V1 Clean RerunsScience16.514.8−1.7
Code MigrationCode47.345.6−1.6
ScBench-Long V0 VerifiedScience25.827.3+1.5
Extended NYT Connections 940 (Quadratic v1)Games91.189.6−1.5
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine16.014.7−1.3
LiveBench 2.0Agents76.878.0+1.3
Finance Agent v2 - ComparablesFinance & Accounting44.946.1+1.2
LMArena Text Arena EloChat & Writing14821475−1.2
ARC-AGI-3 (Semi-Private)General QA1.50.5−1.0
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting66.467.4+1.0
Finance Agent v2 - AdjustmentsFinance & Accounting48.449.2+0.9
ARC-AGI-2General QA72.172.9+0.8
Finance Agent v2 - Earnings AnalysisFinance & Accounting72.172.9+0.8
Artificial Analysis Progress Index v1Indexes318320+0.7
ProgramBenchCode1.00.5−0.5
EpiBench V1 Clean RerunsScience25.825.5−0.3
LMArena Agent Arena Bash RecoveryAgents7.27.9+0.3
Finance Agent v2 - Financial ModelingFinance & Accounting23.323.1−0.3
LegalBenchLegal83.683.3−0.3
Arena Alignment IndexSafety74.574.8+0.3
MedScribeHealth & Medicine85.885.5−0.3
APEX-AccountingFinance & Accounting48.048.1+0.1
AIME 2025Math98.398.30.0
OTIS Mock AIME 2024-2025Math98.398.30.0
ReactBench (39 tasks): Writing ReactCode47.047.00.0

Opus 4.8vsDeepSeek V4.1 Flashnormalised score marginsData Nov 15, 2026Data: Vals AI, Mercor, benchmarks.bio and 42 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…