Atlas

Compare Models

DeepSeek V4.1 Flash vs GPT-5.6 Sol

GPT-5.6 Sol leads by 3.5 capability points (≈95% interval +1.4 to +5.5): a clear lead over DeepSeek V4.1 Flash · P(higher) ≈ 100%

Wins 151 of 198 shared benchmarks (45 lost, 2 tied)

Capability vs. relative cost
DeepSeek V4.1 FlashGPT-5.6 SolClosed 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 FlashvsGPT-5.6 Solcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins45 – 151 · 198 shared, 2 tied

Comparing matching setups only (59 benchmarks excluded) · suites counted once
← DeepSeek V4.1 FlashGPT-5.6 Sol →

DeepSeek V4.1 FlashvsGPT-5.6 Solwins by categoryData Nov 15, 2026Data: Vals AI, benchmarks.bio, Arena and 49 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-5.6 Sol minus DeepSeek V4.1 Flash.

FrontierMath Tier 4 (v2) - PrivateMath26.882.9+56.1
IOI (Vals v2)Code40.391.2+50.9
RiemannBenchMath24.074.4+50.4
Surface Evolver BenchAgents46.393.1+46.9
KernelBench Mega Normalized (RTX PRO 6000)Code48.27.4−40.8
Concept Synth INDUCTION Challenge64 Round 1General QA17.257.8+40.6
KernelBench Hard Normalized (RTX PRO 6000)Code69.731.9−37.8
LLM Creative Story WritingChat & Writing-0.52.5+33.4
ChartographyMultimodal11.645.0+33.4
YC-BenchGames1657236728001−32.4
Concept Synth INDUCTION Challenge100 Round 1General QA11.043.0+32.0
ArXivMath 06/2026Math57.688.5+30.9
SRE Bench — Fully SolvedCode0.830.5+29.8
P2NPCBench (NP-complete, pass@1)Code39.368.4+29.1
ProofBench v1.1Math54.083.0+29.0
BrokenArXiv 06/2026Math39.467.3+27.9
TaxCalcBench TY25Finance & Accounting4.030.0+26.0
Design Creativity BenchCode43.651.3+24.5
A-Fantasia (Valid Attempts with Retries)Games59.336.0+23.3
Furniture Assembly BenchmarkMultimodal34.256.7+22.5
BabyVisionMultimodal43.365.5+22.2
Creative Writing v3 EloChat & Writing15401974+21.9
FrontierMath Tiers 1-3 (v2) - PrivateMath67.489.1+21.8
ALE-BenchCode18292410+20.6
BuseyBench SVGMultimodal4.16.1+20.2
Terminal-Bench Science 0.1 (Vals)Science0.020.0+20.0
GBENCH One-shot Coding GScoreGames48.067.7+19.7
ARC-AGI-2General QA72.992.5+19.6
LLM Chess EloGames10871550+19.0
ReactBench (39 tasks): Fixing ReactCode18.837.5+18.8
ExploitGym - release-reported percentage, cohort unspecifiedCybersecurity15.333.7+18.4
SkillsBench without SkillsAgents61.743.3−18.3
ScBench-Long V0 VerifiedScience27.343.9+16.7
SEC-Bench Pro - release-reported cohortCybersecurity62.879.1+16.3
SkillsBench with SkillsAgents69.854.1−15.7
Excel Modeling BenchmarkFinance & Accounting57.272.3+15.1
Mystery Game PuzzlesGames43.058.0+15.0
ScBench-Long V1 Clean RerunsScience14.829.6+14.8
TxBench Preclinical Pharmacology V1 Clean RerunsScience46.761.3+14.7
SUPERChemScience56.771.0+14.3
CritPt (avg@5)Science14.328.6+14.3
Conceptual Reasoning IndexIndexes58.468.7+14.2
Wiz Cyber Model Arena v3.9Cybersecurity50.264.2+14.0
VIALSMultimodal12.626.5+13.9
AA-Omniscience IndexIndexes-5.322.0+13.6
Terminal-Bench-Science 0.1Science9.022.4+13.3
RareDiseaseBench V1 (35 evaluations)Science0.013.3+13.3
Terminal-Bench 4.0 (Artificial Analysis)Code26.839.9+13.1
GeneBench-Pro Public Release SubsetScience11.024.0+13.0
AA-AnalystAgent (pass^5)Professional Work35.047.5+12.5
MysteryMechanismScience21.233.3+12.2
Biopharma Bench V0.1Health & Medicine57.044.9−12.1
Senior SWE-BenchCode22.734.7+12.0
CadQueryEval (Corrected Geometry)Code76.088.0+12.0
Vals Science IndexIndexes29.641.5+11.8
GameBench 2Games68.480.1+11.6
LMCAGeneral QA47.058.5+11.5
SpatialBench-Long V1 Clean RerunsScience26.237.5+11.3
Humanity's Last Exam DiamondGeneral QA19.130.4+11.3
BrowseComp (Mercor 130-question subset)Search79.490.6+11.2
Chess Puzzles (task version 2.0)Games48.059.0+11.0
LMArena Agent Arena Chat Net Improvement (Steering Burden)Agents1.66.1+11.0
GDP.pdfProfessional Work19.830.7+10.9
Creative Writing v3 Rubric ScoreChat & Writing14.116.2+10.6
Eval Connections One-Shot with TrapsGames73.083.0+10.0
LMArena Agent Arena Work Net Improvement (Steering Burden)Agents2.56.6+9.9
FrontierCUA pass@1Agents6.716.4+9.7
Arena Alignment IndexSafety74.884.2+9.4
ITBench-AAAgents46.956.2+9.3
ORCA-bench Hard RCA (Fireworks 324-task subset)Agents20.329.5+9.3
WeirdML v3Science5.815.0+9.2
LMArena Agent Arena Code Net Improvement (Steering Burden)Agents2.76.7+9.1
AutomationBench 1.0.6 (provider-reported)Agents54.845.8−9.0
BullshitBench v2 LeaderboardSafety57.048.0−9.0
LLM Debate BenchmarkChat & Writing15821645+8.9
AutomationBench-AA 1.0.6Agents68.960.1−8.8
Humanity's Last ExamGeneral QA36.845.5+8.7
Bedside BenchHealth & Medicine82.991.1+8.2
LMArena Agent Arena Net ImprovementAgents3.96.3+7.9
ACCoRDGeneral QA68.175.9+7.8
Longform Creative WritingChat & Writing74.081.7+7.7
Artificial Analysis Intelligence Index v4.3.2Indexes39.547.0+7.5
Artificial Analysis Intelligence Index v4.3Indexes39.547.1+7.5
OpenDS-WorldSearch20.528.0+7.5
MacroscopeBenchCode70.577.8+7.3
Code MigrationCode45.652.9+7.3
NL2Repo - release-reported repository generation scoreCode64.056.8−7.2
AA-LCR v1.1Search55.362.3+7.0
ARC-AGI-3 (Semi-Private)General QA0.97.8+6.9
BioSecBench-Structure V1 (58 evaluations)Science27.620.7−6.9
Legal Research BenchLegal41.348.1+6.7
MetagenomicsBench V0Science38.031.3−6.7
MetagenomicsBench V1Science38.031.3−6.7
HealthBenchHealth & Medicine61.955.3−6.6
Nimble Web ContextSearch73.580.0+6.5
Vision Arena OverallMultimodal12631288+6.4
MMMU ProMultimodal65.671.9+6.3
NeuroscienceBench V1 (91 evaluations)Science22.015.8−6.2
DistressBenchSafety79.873.6−6.2
AutoformBenchMath52.858.9+6.2
Terminal-Bench 4.0Code31.237.3+6.1
ReactBench (39 tasks): Writing ReactCode47.053.0+6.1
LMArena Agent Arena Chat Praise vs ComplaintAgents3.115.0+6.0
SciCode 1.0.1 (Artificial Analysis)Code51.957.8+5.9
DTBenchGeneral QA89.995.7+5.9
Vals Index v2Indexes57.963.7+5.8
LMArena Agent Arena Code Bash ReliabilityAgents-0.211.5+5.8
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting69.875.6+5.8
LMArena Agent Arena Work Bash ReliabilityAgents0.111.6+5.7
Tax Agent BenchFinance & Accounting62.568.0+5.5
LMArena Agent Arena Chat Bash ReliabilityAgents-1.09.7+5.4
BioSecBench-Function V1 Clean RerunsScience40.835.5−5.3
Epoch Capabilities Index (ECI)Indexes156163+5.3
TxBench Antibody Discovery V0Science39.033.8−5.2
Harvey LAB held-out set (Fireworks run)Legal5.70.6−5.1
ScBench V1 Clean RerunsScience54.759.8+5.1
Agents' Last Exam - release-reported pass rate, subset unspecifiedAgents31.826.7−5.1
SciCode (main problems)Science11.216.2+5.0
SpatialBench V1 Clean RerunsScience63.268.1+4.9
LMArena Agent Arena SteerabilityAgents-3.36.5+4.9
EnterpriseBench: CoreCraftProfessional Work41.546.2+4.7
Terminal-Bench 2.1Code84.188.8+4.7
SAGEMath47.952.6+4.7
LMArena Agent Arena Code Steering BurdenAgents2.311.4+4.5
TxBench-Antibody-Development V1 (60 evaluations)Science14.418.9+4.5
Terminal-Bench 3 - release-reported accuracyCode30.034.4+4.4
Vibe Code Bench v1.1Code84.780.5−4.2
VariantBench V1 Clean RerunsScience31.135.3+4.2
BioSecBench-Surveillance V1 Clean RerunsScience51.547.3−4.2
Extended NYT Connections 940 (Quadratic v1)Games89.693.8+4.2
Harvey's Legal Agent BenchmarkLegal6.72.5−4.2
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine14.718.7+4.0
FrontierCode 1.1 Extended - ScoreCode56.660.6+4.0
Artificial Analysis Progress Index v1Indexes320331+3.9
FoodTruck BenchGames5017953229+3.9
APEX-SWECode51.948.0−3.9
LMArena Agent Arena Work Praise vs ComplaintAgents3.711.5+3.9
LMArena Agent Arena Work Steering BurdenAgents1.59.2+3.8
LegalBenchLegal83.387.0+3.7
BigFinanceBench private holdout (final-answer accuracy)Finance & Accounting40.644.2+3.6
CyberGymCybersecurity88.184.5−3.6
LMArena Agent Arena Bash RecoveryAgents7.90.7−3.6
APEX-AccountingFinance & Accounting48.151.5+3.4
Medical Long Context Reasoning (MLCR-AA)Health & Medicine22.826.1+3.3
BioMysteryBench (Vals Terminus 2)Science67.871.1+3.3
ARC-AGI-1 Semi-PrivateGeneral QA94.597.5+3.0
AA-Briefcase EloAgents14241475+3.0
LMArena Agent Arena Code Praise vs ComplaintAgents3.08.7+2.8
LMArena Agent Arena Chat Steering BurdenAgents-0.55.2+2.8
MedCodeHealth & Medicine41.244.0+2.8
Political Science LLM BenchmarkProfessional Work66.569.2+2.7
CyberBench v1.1 OverallCybersecurity73.776.3+2.6
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting67.464.9−2.5
Dual Exam Agent IndexIndexes18.320.7+2.4
TxBench Oligo Discovery V1 (113 evaluations)Science34.837.2+2.4
Finance Agent v2 - Financial ModelingFinance & Accounting23.120.7−2.3
TxBench Antibody Discovery V1 (100 evaluations)Science39.737.3−2.3
GDPval-AA v2Professional Work16321585−2.3
dfbench Detection RecallCybersecurity57.359.6+2.3
APEX-Agents 1.1 Corporate Law (68-task image-free slice)Legal78.480.6+2.2
Public Benefits Bench v1.1Professional Work64.366.5+2.2
Financial Coaching BenchFinance & Accounting72.874.9+2.1
Finance Agent v2 - PrecedentsFinance & Accounting28.226.2−2.0
Finance Agent v2 - Disclosure AnalysisFinance & Accounting61.363.3+2.0
LiveBench 2.0Agents78.076.3−1.8
LMArena Text Arena EloChat & Writing14751485+1.7
DeepResearch Bench II (Mercor 130-task set)Search53.251.5−1.7
SpeciEval (Refusal-Aware Retries, Historical)Safety97.499.0+1.7
Harvey LAB: In-House Contracting (held-out)Legal0.32.0+1.7
OTIS Mock AIME 2024-2025Math98.3100.0+1.7
SpeciEval (Pooled English Runs)Safety97.499.0+1.7
SpatialBench-Long V0 VerifiedScience34.936.4+1.5
Finance Agent v2 - Earnings AnalysisFinance & Accounting72.974.4+1.5
LMArena Agent Arena Work Confirmed SuccessAgents5.72.8−1.4
LMArena Agent Arena Code Confirmed SuccessAgents6.63.9−1.3
LMArena Agent Arena Chat Confirmed SuccessAgents4.92.4−1.3
ProgramBenchCode0.51.5+1.0
Terminal-Bench 2.1 Extended (Mercor held-out)Code38.437.4−1.0
SpatialBench V0 VerifiedScience75.474.5−0.9
GDPval-AA v2.1Professional Work16001619+0.8
Finance Agent v2 - Market AnalysisFinance & Accounting63.464.2+0.8
ComplexConstraintsChat & Writing49.750.5+0.8
Finance Agent v2 - ComparablesFinance & Accounting46.145.3−0.8
TissueMicroscopyBench V1 (49 evaluations)Science2.72.0−0.7
BioSecBench-RespiratoryVirus-Surveillance V1 (110 evaluations)Science5.25.8+0.6
TxBench Oligo Discovery V0Science33.934.5+0.6
BioSecBench-Function V0Science38.738.3−0.4
EpiBench V1 Clean RerunsScience25.525.8+0.3
Finance Agent v2 - All PassFinance & Accounting41.141.4+0.3
MedScribeHealth & Medicine85.585.2−0.3
DeepSWE v1.1Code72.672.7+0.1
τ³-Banking (Fireworks, banking_knowledge)Agents35.735.7−0.1
LMArena Agent Arena Chat Tool HallucinationAgents0.30.2−0.1
LMArena Agent Arena Code Tool HallucinationAgents0.50.6+0.1
LMArena Agent Arena Work Tool HallucinationAgents0.50.60.0
Finance Agent v2 - AdjustmentsFinance & Accounting49.249.20.0
EpiBenchScience42.142.10.0
SWE Atlas - Codebase QnACode53.553.50.0

DeepSeek V4.1 FlashvsGPT-5.6 Solnormalised score marginsData Nov 15, 2026Data: Vals AI, benchmarks.bio, Arena and 49 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…