Atlas

Compare Models

DeepSeek V4.1 Flash vs GPT-6 Sol

GPT-6 Sol leads by 4.4 capability points (≈95% interval +2.3 to +6.5): a clear lead over DeepSeek V4.1 Flash · P(higher) ≈ 100%

Wins 136 of 188 shared benchmarks (52 lost)

Capability vs. relative cost
DeepSeek V4.1 FlashGPT-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-6 Solcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins52 – 136 · 188 shared

Comparing matching setups only (72 benchmarks excluded) · suites counted once
← DeepSeek V4.1 FlashGPT-6 Sol →

DeepSeek V4.1 FlashvsGPT-6 Solwins by categoryData Nov 15, 2026Data: Artificial Analysis, Mercor, Arena and 38 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6 Sol minus DeepSeek V4.1 Flash.

AA-PersonalityBench Organization indexChat & Writing10012.4−88.5
AA-PersonalityBench Informativeness indexChat & Writing79.110.5−69.2
FrontierMath Tier 4 (v2) - PrivateMath26.890.0+63.2
BrokenArXiv 05/2026Math28.587.0+58.5
BrokenArXiv 06/2026Math39.495.4+56.0
ArXivMath 08/2026 (Native Harnesses)Math43.993.0+49.1
AA-PersonalityBench Conviction indexChat & Writing96.250.5−46.2
KernelBench Mega Normalized (RTX PRO 6000)Code48.22.7−45.5
SWE-bench Verified Extended (Mercor held-out)Code82.739.3−43.4
IOI (Vals v2)Code40.382.6+42.3
ChartographyMultimodal11.653.6+42.0
RiemannBenchMath24.064.8+40.8
AA-PersonalityBench Slop indexChat & Writing37.21.0−36.5
BuseyBench SVGMultimodal4.17.5+34.4
ArXivMath 06/2026Math57.691.0+33.3
ArXivMath 05/2026Math58.390.0+31.7
Terminal-Bench Science 0.1 (Vals)Science0.030.0+30.0
SRE Bench — Fully SolvedCode0.830.5+29.8
Creative Writing v3 EloChat & Writing15402126+29.6
ProofBench v1.1Math54.083.0+29.0
AA-PersonalityBench Expressiveness indexChat & Writing61.933.4−28.8
ALE-BenchCode18292637+28.6
MMMU-Pro Extended (Mercor held-out)Multimodal25.051.9+26.9
BioSecBench-Refusal V1 Clean RerunsSafety11.337.9+26.6
Humanity's Last Exam DiamondGeneral QA19.145.2+26.1
Furniture Assembly BenchmarkMultimodal34.258.3+24.2
TaxCalcBench TY25Finance & Accounting4.028.0+24.0
CadQueryEval (Corrected Geometry)Code76.0100.0+24.0
BrokenArXiv 08/2026 (Native Harnesses)Math54.878.6+23.8
CyberGym-E2E-AA (Artificial Analysis)Cybersecurity22.90.0−22.9
FrontierMath Tiers 1-3 (v2) - PrivateMath67.489.8+22.5
GBENCH One-shot Coding GScoreGames48.070.1+22.1
AutomationBench 1.0.6 (provider-reported)Agents54.833.2−21.6
AA-PersonalityBench Transparency indexChat & Writing60.039.1−21.2
Terminal-Bench-Science 0.1Science9.030.0+21.0
ARC-AGI-3 (Semi-Private)General QA2.723.0+20.3
MedXpertQA MM Extended (Mercor held-out)Health & Medicine48.468.0+19.6
Wiz Cyber Model Arena v3.9Cybersecurity50.269.3+19.1
LLM Chess EloGames10871545+18.8
APEX-Agents 1.1Professional Work50.568.7+18.2
BizAgentBench (strict success)Agents43.361.3+18.0
MedXpertQA MM (200-question set)Health & Medicine63.581.3+17.8
SimpleQA Verified (Epoch anti-abstention)General QA42.960.7+17.8
LMArena Agent Arena Code Net Improvement (Steering Burden)Agents2.710.6+17.8
Conceptual Reasoning IndexIndexes58.471.0+17.3
Terminal-Bench 4.0 (Artificial Analysis)Code26.843.9+17.2
ARC-AGI-2General QA72.989.6+16.7
CritPt (avg@5)Science14.330.9+16.6
SWE-bench MultilingualCode98.281.7−16.5
TxBench Preclinical Pharmacology V1 Clean RerunsScience46.763.0+16.3
LMArena Agent Arena Net ImprovementAgents3.98.8+16.3
AA-Omniscience IndexIndexes-5.327.1+16.2
SUPERChemScience56.772.8+16.1
Eval Connections One-Shot with TrapsGames73.089.0+16.0
Vals Science IndexIndexes29.645.0+15.3
DeepsecBench-AA (Artificial Analysis)Cybersecurity31.246.3+15.1
GeneBench-Pro Public Release SubsetScience11.026.0+15.0
CharXiv Extended (Mercor held-out)Multimodal68.983.8+14.9
LMArena Agent Arena Work Net Improvement (Steering Burden)Agents2.58.6+14.7
Quant-Arena (iterative total score)Finance & Accounting36.551.1+14.7
Excel Modeling BenchmarkFinance & Accounting57.271.5+14.3
ACCoRDGeneral QA68.182.4+14.3
WeirdML v3Science5.819.7+13.9
BioSecBench-Structure V1 (58 evaluations)Science27.613.8−13.8
ScBench-Long V1 Clean RerunsScience14.828.4+13.6
HealthBenchHealth & Medicine61.948.2−13.6
AA-PersonalityBench Warmth indexChat & Writing41.054.3+13.5
FrontierCUA pass@1Agents6.720.0+13.3
Mystery Game PuzzlesGames43.056.0+13.0
Arena Alignment IndexSafety74.887.6+12.8
Terminal-Bench 4.0Code31.243.8+12.6
Legal Research BenchLegal41.328.8−12.5
GDP.pdf (Artificial Analysis, All-pass)Professional Work12.825.2+12.4
LMCAGeneral QA47.059.1+12.1
DistressBenchSafety79.868.0−11.8
Code MigrationCode45.657.2+11.6
τ³-Banking (Fireworks, banking_knowledge)Agents35.747.0+11.3
Decision-Bench v1.1Professional Work25.336.5+11.2
GameBench 2Games68.479.4+11.0
LMArena Agent Arena Chat Net Improvement (Steering Burden)Agents1.65.9+10.4
DeepResearch Bench II (Mercor 130-task set)Search53.242.9−10.3
Financial Coaching BenchFinance & Accounting72.862.6−10.2
BrowseComp (Mercor 130-question subset)Search79.489.4+10.0
Creative Writing v3 Rubric ScoreChat & Writing14.116.1+10.0
APEX-Accounting 1.1Finance & Accounting50.159.7+9.6
Tax Agent BenchFinance & Accounting62.553.0−9.4
SWE Atlas - Codebase QnACode53.544.1−9.4
MysteryMechanismScience21.230.2+9.0
LMArena Agent Arena SteerabilityAgents-3.314.5+8.9
Longform Creative WritingChat & Writing74.082.8+8.8
AA-LCR v1.1Search55.364.0+8.7
MacroscopeBenchCode72.080.3+8.3
HealthBench ProfessionalHealth & Medicine52.961.2+8.3
TxBench Oligo Discovery V1 (113 evaluations)Science34.843.1+8.3
Artificial Analysis Intelligence Index v4.3.2Indexes39.547.6+8.2
Chess Puzzles (task version 2.0)Games48.056.0+8.0
DualEntry Accounting AI BenchmarkFinance & Accounting84.276.2−8.0
Public Benefits Bench v1.1Professional Work64.356.6−7.6
VariantBench V1 Clean RerunsScience31.138.7+7.6
Humanity's Last Exam (Text-Only)General QA10.818.4+7.6
YC-BenchGames16572361441793−7.5
DTBenchGeneral QA89.997.3+7.5
LMArena Agent Arena Chat Bash ReliabilityAgents-1.013.7+7.4
AutomationBench-AA 1.0.6Agents68.961.7−7.2
BioMysteryBench (Vals Terminus 2)Science67.874.8+7.0
APEX-SWECode51.945.0−6.9
ScBench V1 Clean RerunsScience54.761.5+6.8
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine60.954.2−6.7
LMArena Agent Arena Code Steering BurdenAgents2.315.7+6.7
Medical Long Context Reasoning (MLCR-AA)Health & Medicine22.816.1−6.7
BigFinanceBench private holdout (final-answer accuracy)Finance & Accounting40.647.1+6.5
SimpleBenchGeneral QA66.773.1+6.4
Epoch Capabilities Index (ECI)Indexes156164+6.1
DeepSWE v1.1Code72.666.6−6.0
MedCodeHealth & Medicine41.247.1+5.9
SciCode 1.0.1 (Artificial Analysis)Code51.957.6+5.8
LMArena Agent Arena Work Steering BurdenAgents1.513.0+5.7
SWE Atlas Codebase QnA Extended (Mercor held-out)Code36.731.3−5.4
LMArena Agent Arena Code Praise vs ComplaintAgents3.013.8+5.4
SpatialBench-Long V1 Clean RerunsScience26.231.6+5.4
LMArena Agent Arena Work Bash ReliabilityAgents0.110.8+5.3
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine14.720.0+5.3
LMArena Agent Arena Code Bash ReliabilityAgents-0.210.3+5.2
τ³-Banking (Mercor run, bm25_grep retrieval)Agents37.843.0+5.2
Bedside BenchHealth & Medicine82.988.1+5.2
Terminal-Bench 2.1Code90.685.4−5.2
LMArena Agent Arena Work Praise vs ComplaintAgents3.713.8+5.1
Harvey LAB held-out set (Fireworks run)Legal5.70.7−5.0
Harvey's Legal Agent BenchmarkLegal6.71.7−5.0
KernelBench CUDA Normalized (RTX PRO 6000)Code47.042.0−4.9
CharXiv (descriptive and reasoning combined)Multimodal90.195.0+4.9
Artificial Analysis Progress Index v1Indexes320334+4.8
MMMU ProMultimodal65.670.3+4.7
Vals Index v2Indexes57.962.6+4.7
Finance Agent v2Finance & Accounting53.549.0−4.4
SpatialBench V1 Clean RerunsScience63.267.5+4.4
CyberBench v1.1 OverallCybersecurity73.778.0+4.3
CWE-Bench-AA (Artificial Analysis)Cybersecurity68.364.2−4.2
MetagenomicsBench V0Science38.042.0+4.0
MetagenomicsBench V1Science38.042.0+4.0
Terminal-Bench 2.1 Extended (Mercor held-out)Code38.442.4+4.0
Artificial Analysis Cyber Index v1Indexes40.836.8−4.0
LiveBench 2.0Agents78.081.9+3.9
Design Arena Overall Code CategoriesCode13161293−3.8
GDPval-AA v2.1Professional Work16001514−3.8
BioSecBench-Surveillance V1 Clean RerunsScience51.547.7−3.7
Epiq LAER CorporateBenchProfessional Work75.379.0+3.7
TxBench Antibody Discovery V1 (100 evaluations)Science39.743.3+3.7
MedScribeHealth & Medicine85.582.0−3.5
EpiBench V1 Clean RerunsScience25.522.0−3.5
AA-Briefcase v1.1 EloAgents14151476+3.3
GDPval (Mercor 213-task rubric score)Professional Work86.783.5−3.2
LMArena Text Arena EloChat & Writing14751457−3.2
AdvancedIFChat & Writing84.080.9−3.1
SAGEMath47.944.8−3.1
Vibe Code Bench v1.1Code84.787.8+3.1
Nimble Web ContextSearch73.576.5+3.0
BullshitBench v2 LeaderboardSafety57.054.0−3.0
LMArena Agent Arena Chat Steering BurdenAgents-0.55.5+3.0
LMArena Agent Arena Code Confirmed SuccessAgents6.612.3+2.9
Biopharma Bench V0.1Health & Medicine57.054.3−2.7
SWE-bench VerifiedCode82.980.3−2.6
ITBench-AAAgents46.949.4+2.5
Harvey LAB-AA v1.1 (hallucination-gated)Legal1.73.6+1.9
AA-PersonalityBench Imagination indexChat & Writing31.529.6−1.9
APEX-1 (Mercor, mean score)Professional Work65.563.6−1.9
AIME 2025Math98.3100.0+1.7
LMArena Agent Arena Chat Confirmed SuccessAgents4.98.3+1.7
OTIS Mock AIME 2024-2025Math98.3100.0+1.7
EnterpriseBench: CoreCraftProfessional Work41.543.1+1.6
BioSecBench-Function V1 Clean RerunsScience40.839.3−1.6
ProgramBenchCode0.52.0+1.5
SciCode (main problems)Science11.212.7+1.5
LMArena Agent Arena Chat Praise vs ComplaintAgents3.15.5+1.2
SpeciEval (Pooled English Runs)Safety97.498.5+1.1
SpeciEval (Refusal-Aware Retries, Historical)Safety97.498.5+1.1
ComplexConstraintsChat & Writing49.750.8+1.1
ARC-AGI-1 Semi-PrivateGeneral QA94.595.5+1.0
Harvey LAB: In-House Contracting (held-out)Legal0.31.3+1.0
MMLU ProGeneral QA88.288.9+0.7
AA-LCR Extended (Mercor held-out)Search70.870.2−0.6
Extended NYT Connections 940 (Quadratic v1)Games89.690.1+0.5
LMArena Agent Arena Bash RecoveryAgents7.97.0−0.4
dfbench Detection RecallCybersecurity57.357.7+0.4
LMArena Agent Arena Work Confirmed SuccessAgents5.76.0+0.2
LMArena Agent Arena Code Tool HallucinationAgents0.50.7+0.1
LMArena Agent Arena Work Tool HallucinationAgents0.50.7+0.1
LMArena Agent Arena Chat Tool HallucinationAgents0.30.40.0

DeepSeek V4.1 FlashvsGPT-6 Solnormalised score marginsData Nov 15, 2026Data: Artificial Analysis, Mercor, Arena and 38 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…