Atlas

Compare Models

Claude Sonnet 5.5 vs DeepSeek V4.1 Flash

Claude Sonnet 5.5 leads by 7.2 capability points (≈95% interval +4.9 to +9.5): a clear lead over DeepSeek V4.1 Flash · P(higher) ≈ 100%

Wins 152 of 179 shared benchmarks (27 lost)

Capability vs. relative cost
Claude Sonnet 5.5DeepSeek 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.

Claude Sonnet 5.5vsDeepSeek V4.1 Flashcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins152 – 27 · 179 shared

Comparing matching setups only (15 benchmarks excluded) · suites counted once
← Claude Sonnet 5.5DeepSeek V4.1 Flash →

Claude Sonnet 5.5vsDeepSeek V4.1 Flashwins by categoryData Nov 15, 2026Data: Vals AI, Artificial Analysis, Mercor and 41 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 Claude Sonnet 5.5.

AA-PersonalityBench Imagination indexChat & Writing84.831.5−53.8
FrontierMath Tier 4 (v2) - PrivateMath80.526.8−53.7
Medical Long Context Reasoning (MLCR-AA)Health & Medicine75.022.8−52.2
ChartographyMultimodal61.611.6−50.0
Terminal-Bench Science 0.1 (Vals)Science47.10.0−47.1
ProofBench v1.1Math100.054.0−46.0
Terminal-Bench-Science 0.1Science53.39.0−44.3
IOI (Vals v2)Code83.140.3−42.8
KernelBench Mega Normalized (RTX PRO 6000)Code90.648.2−42.4
Furniture Assembly BenchmarkMultimodal75.034.2−40.8
AA-PersonalityBench Transparency indexChat & Writing10060.0−40.4
KernelBench CUDA Normalized (RTX PRO 6000)Code87.047.0−40.1
AA-PersonalityBench Conviction indexChat & Writing58.196.2+38.5
LMArena Agent Arena Chat Net Improvement (Steering Burden)Agents17.11.6−37.7
Terminal-Bench 4.0 (Artificial Analysis)Code63.626.8−36.9
Humanity's Last Exam DiamondGeneral QA53.319.1−34.2
APEX-Agents 1.1Professional Work83.750.5−33.2
Concept Synth INDUCTION Challenge100 Round 1General QA44.011.0−33.0
Concept Synth INDUCTION Challenge64 Round 1General QA50.017.2−32.8
AA-PersonalityBench Warmth indexChat & Writing73.341.0−32.7
ARC-AGI-3 (Semi-Private)General QA35.53.2−32.3
Code2Video-AA EloCode14431053−31.7
Terminal-Bench 4.0Code61.831.2−30.6
LMArena Agent Arena Code Net Improvement (Steering Burden)Agents15.82.7−29.5
EnterpriseBench: CoreCraftProfessional Work70.841.5−29.3
Arena Image-to-WebDevCode17441610−29.1
VIALSMultimodal41.012.6−28.4
Bug Hunt BenchCode48.920.6−28.3
GeneBench-Pro Public Release SubsetScience39.011.0−28.0
Code Migration v1.1Code86.958.9−27.9
MysteryMechanismScience49.121.2−27.9
LMArena Agent Arena Work Net Improvement (Steering Burden)Agents13.72.5−27.0
OpenWorldEngineer (Mean@8)Code39.012.4−26.6
Decision-Bench v1.1Professional Work51.225.3−25.9
Conceptual Reasoning IndexIndexes77.158.4−25.7
SWE-bench Verified Extended (Mercor held-out)Code57.382.7+25.4
BuseyBench SVGMultimodal6.54.1−24.4
Code MigrationCode69.845.6−24.2
GameBench 2Games91.468.4−23.0
AA-AnalystAgent (pass^5)Professional Work57.535.0−22.5
TxBench-Antibody-Development V1 (60 evaluations)Science36.714.4−22.2
MetagenomicsBench V0Science60.038.0−22.0
MetagenomicsBench V1Science60.038.0−22.0
Mystery Game PuzzlesGames65.043.0−22.0
AA-Briefcase v1.1 EloAgents18161415−22.0
ALE-BenchCode24481829−21.9
FrontierMath Tiers 1-3 (v2) - PrivateMath88.867.4−21.4
LMCAGeneral QA68.347.0−21.3
VariantBench V1 Clean RerunsScience51.731.1−20.6
Humanity's Last ExamGeneral QA56.936.8−20.1
LMArena Code Arena WebDevCode17731621−19.9
GBENCH One-shot Coding GScoreGames67.448.0−19.4
AA-Omniscience IndexIndexes32.3-5.3−18.8
Excel Modeling BenchmarkFinance & Accounting75.757.2−18.5
HealthBench ProfessionalHealth & Medicine70.852.9−17.9
TxBench Preclinical Pharmacology V1 Clean RerunsScience64.346.7−17.7
ACCoRDGeneral QA85.768.1−17.6
MedXpertQA MM (200-question set)Health & Medicine80.763.5−17.2
CritPt (avg@5)Science31.414.3−17.1
TxBench Oligo Discovery V1 (113 evaluations)Science51.934.8−17.1
LMArena Agent Arena Chat Praise vs ComplaintAgents37.33.1−17.1
Artificial Analysis Intelligence Index v4.3.2Indexes56.039.5−16.5
τ³-Banking (Mercor run, bm25_grep retrieval)Agents54.037.8−16.2
ScBench-Long V1 Clean RerunsScience30.314.8−15.5
AA-PersonalityBench Organization indexChat & Writing84.8100+15.4
SAFE-Teen 106 worst branch: Urgent Safety (under-18 prompt)Safety11.827.2−15.4
CharXiv Extended (Mercor held-out)Multimodal84.268.9−15.3
TxBench Antibody Discovery V1 (100 evaluations)Science54.739.7−15.0
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine29.314.7−14.7
APEX-SWECode66.451.9−14.5
APEX-Accounting 1.1Finance & Accounting64.550.1−14.4
SAFE-Teen 106 worst branch: Relationships (no prompt)Safety37.051.1−14.2
TaxCalcBench TY25Finance & Accounting18.04.0−14.0
WeirdML v3Science19.95.9−14.0
SWE-bench VerifiedCode96.782.9−13.8
RuneBenchGames4.85.7+13.2
SAFE-Teen 106 worst branch: Relationships (under-18 prompt)Safety7.319.8−12.5
BioSecBench-RespiratoryVirus-Surveillance V1 (110 evaluations)Science17.35.2−12.1
SWE-bench Multimodal (Vals v1, 480 Tasks)Code58.346.5−11.9
MedCodeHealth & Medicine52.941.2−11.7
GDP.pdfProfessional Work31.419.8−11.6
APEX-1 (Mercor, mean score)Professional Work53.965.5+11.6
AA-PersonalityBench Expressiveness indexChat & Writing73.361.9−11.5
Chess Puzzles (task version 2.0)Games59.048.0−11.0
ARC-AGI-2General QA83.972.9−11.0
BioSecBench-Function V1 Clean RerunsScience51.740.8−10.9
LMArena Agent Arena Code Praise vs ComplaintAgents24.53.0−10.7
MineBench — Bradley-Terry RatingGames19981822−10.7
DAXBench v2Code86.275.5−10.6
Artificial Analysis Progress Index v1Indexes351320−10.6
GDPval-AA v2.1Professional Work18381600−10.5
SAFE-Teen 106: Relationships (no prompt)Safety16.927.1−10.2
LMArena Agent Arena Chat Confirmed SuccessAgents24.74.9−9.9
AA-PersonalityBench Slop indexChat & Writing27.737.2+9.6
SpatialBench V1 Clean RerunsScience72.863.2−9.6
SimpleBenchGeneral QA75.966.7−9.2
SciCode 1.0.1 (Artificial Analysis)Code61.051.9−9.1
Extended NYT Connections 940 (Quadratic v1)Games80.589.6+9.1
SpatialBench-Long V1 Clean RerunsScience34.926.2−8.7
LMArena Agent Arena Code Confirmed SuccessAgents23.86.6−8.6
LMArena Agent Arena Work Praise vs ComplaintAgents20.63.7−8.5
LMArena Agent Arena Work Confirmed SuccessAgents22.15.7−8.2
SAFE-Teen 106 worst branch: Urgent Safety (no prompt)Safety23.231.4−8.2
Excel Modeling Benchmark v1.1Finance & Accounting75.967.8−8.1
SWE Atlas Codebase QnA Extended (Mercor held-out)Code44.736.7−8.0
Epoch Capabilities Index (ECI)Indexes166156−8.0
FrontierCode 1.1 Extended - Pass RateCode69.962.0−7.8
FrontierCode 1.1 Extended - ScoreCode64.456.6−7.8
Vibe Code Bench v1.1Code92.484.7−7.7
LMArena Agent Arena Chat Bash ReliabilityAgents13.9-1.0−7.5
DTBenchGeneral QA97.389.9−7.5
SWE-bench MultilingualCode90.998.2+7.3
MedXpertQA MM Extended (Mercor held-out)Health & Medicine55.648.4−7.2
BrowseComp (Mercor 130-question subset)Search86.679.4−7.2
Terminal-Bench 2.1 Extended (Mercor held-out)Code45.538.4−7.1
LMArena Agent Arena Code Steering BurdenAgents16.22.3−7.0
SAFE-Teen 106: Health Advice (under-18 prompt)Safety16.923.6−6.7
SpeciEval (Refusal-Aware Retries, Historical)Safety91.097.4+6.4
SpeciEval (Pooled English Runs)Safety91.097.4+6.4
SPU (Science Prediction Universe)Science0.30.2−6.3
SAFE-Teen 106 worst branch: Health Advice (under-18 prompt)Safety30.837.0−6.3
Nimble Web ContextSearch67.273.5+6.3
LMArena Agent Arena Work Bash ReliabilityAgents12.50.1−6.2
CharXiv (descriptive and reasoning combined)Multimodal96.190.1−6.0
ProgramBenchCode6.50.5−6.0
LMArena Agent Arena Code Bash ReliabilityAgents11.8-0.2−6.0
AA-PersonalityBench Informativeness indexChat & Writing84.879.1−5.8
BioSecBench-Structure V1 (58 evaluations)Science21.827.6+5.7
ScBench V1 Clean RerunsScience60.354.7−5.6
MedScribeHealth & Medicine91.185.5−5.6
Design Arena Overall Code CategoriesCode13491316−5.5
DistressBenchSafety74.679.8+5.2
LMArena Agent Arena Work Steering BurdenAgents11.91.5−5.2
SAFE-Teen 106: Urgent Safety (under-18 prompt)Safety2.97.9−5.0
EpiBench V1 Clean RerunsScience30.225.5−4.7
Quant-Arena (iterative total score)Finance & Accounting40.936.5−4.5
LiveBench 2.0Agents82.578.0−4.4
LLM Chess EloGames11951087−4.4
DualEntry Accounting AI BenchmarkFinance & Accounting80.284.2+4.0
HealthBenchHealth & Medicine65.961.9−4.0
LMArena Agent Arena Chat Steering BurdenAgents7.4-0.5−3.9
SAGEMath51.847.9−3.9
Harvey's Legal Agent BenchmarkLegal2.96.7+3.7
SimpleQA Verified (Epoch anti-abstention)General QA46.542.9−3.6
LMArena Agent Arena Bash RecoveryAgents15.07.9−3.6
SAFE-Teen 106: Urgent Safety (no prompt)Safety14.711.3+3.4
Immigration BenchLegal69.065.6−3.3
Terminal-Bench 2.1Code87.390.6+3.3
SWE Atlas - Codebase QnACode50.353.5+3.2
SAFE-Teen 106: Health Advice (no prompt)Safety33.230.2+3.0
BullshitBench v2 LeaderboardSafety60.057.0−3.0
AA-LCR Extended (Mercor held-out)Search73.870.8−3.0
Eval Connections One-Shot with TrapsGames76.073.0−3.0
YC-BenchGames17413151657236−2.9
Public Benefits Bench v1.1Professional Work67.264.3−2.9
AutomationBench-AA 1.0.6Agents71.868.9−2.9
MMLU ProGeneral QA91.088.2−2.8
SAGE v1.1Math88.085.4−2.6
SAFE-Teen 106: Relationships (under-18 prompt)Safety1.74.2−2.5
Financial Coaching BenchFinance & Accounting74.972.8−2.1
DeepSWE v1.1Code70.572.6+2.1
Access to Justice BenchmarkLegal74.972.9−2.1
RCM-Bench-1Health & Medicine91.189.0−2.1
ARC-AGI-1 Semi-PrivateGeneral QA96.594.5−2.0
SAFE-Teen 106 worst branch: Health Advice (no prompt)Safety42.941.1+1.8
AIME 2025Math100.098.3−1.7
OTIS Mock AIME 2024-2025Math100.098.3−1.7
Ramp Accounting BenchFinance & Accounting42.540.8−1.7
DeepResearch Bench II (Mercor 130-task set)Search51.753.2+1.5
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine62.160.9−1.2
Harvey LAB-AA v1.1 (hallucination-gated)Legal2.81.7−1.1
BigFinanceBench private holdout (final-answer accuracy)Finance & Accounting41.540.6−0.9
AdvancedIFChat & Writing83.484.0+0.6
LMArena Text Arena EloChat & Writing14721475+0.4
GDPval (Mercor 213-task rubric score)Professional Work86.586.7+0.2
LMArena Agent Arena Chat Tool HallucinationAgents0.50.3−0.1
Vision Arena OverallMultimodal126312630.0
LMArena Agent Arena Work Tool HallucinationAgents0.60.50.0
LMArena Agent Arena Code Tool HallucinationAgents0.50.50.0

Claude Sonnet 5.5vsDeepSeek V4.1 Flashnormalised score marginsData Nov 15, 2026Data: Vals AI, Artificial Analysis, Mercor and 41 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…