Atlas

Compare Models

Claude Sonnet 5.5 vs GPT-5.5

Claude Sonnet 5.5 leads by 6.6 capability points (≈95% interval +4.3 to +8.9): a clear lead over GPT-5.5 · P(higher) ≈ 100%

Wins 107 of 135 shared benchmarks (25 lost, 3 tied)

Capability vs. relative cost
Claude Sonnet 5.5GPT-5.5Closed 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 5.5vsGPT-5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins107 – 25 · 135 shared, 3 tied

Comparing matching setups only (40 benchmarks excluded) · suites counted once
← Claude Sonnet 5.5GPT-5.5 →

Claude Sonnet 5.5vsGPT-5.5wins by categoryData Nov 15, 2026Data: Mercor, Artificial Analysis, benchmarks.bio and 39 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-5.5 minus Claude Sonnet 5.5.

KernelBench Mega Normalized (RTX PRO 6000)Code90.612.2−78.4
Medical Long Context Reasoning (MLCR-AA)Health & Medicine75.017.8−57.2
Terminal-Bench 4.0 (Artificial Analysis)Code57.114.6−42.4
Code2Video-AA EloCode14431000−36.0
ChessBench CoherenceGames64.6100.0+35.4
AA-Briefcase v1.1 EloAgents17401135−33.2
GameBench 2Games91.459.9−31.5
GeneBench-Pro Public Release SubsetScience39.08.0−31.0
Furniture Assembly BenchmarkMultimodal75.044.2−30.8
ActiveVisionMultimodal41.210.6−30.6
ChartographyMultimodal61.631.0−30.6
APEX-SWECode66.436.6−29.8
ChessBench EloGames8571349+29.3
SWE Atlas Codebase QnA Extended (Mercor held-out)Code44.717.3−27.4
Humanity's Last Exam DiamondGeneral QA53.327.9−25.4
Code MigrationCode69.845.2−24.7
AutomationBench-AA 1.0.6Agents71.847.3−24.5
BioSecBench-Function V1 Clean RerunsScience51.728.3−23.4
Vibe Code Bench v1.1Code92.469.8−22.5
Vending-Bench 2Games111657524−22.0
TxBench Antibody Discovery V1 (100 evaluations)Science54.733.0−21.7
VariantBench V1 Clean RerunsScience51.730.5−21.2
SM BenchSafety85.465.4−20.0
EnterpriseBench: CoreCraftProfessional Work70.851.3−19.5
YC-BenchGames17413151206045−18.7
SWE-bench Verified Extended (Mercor held-out)Code57.339.3−18.0
TxBench Oligo Discovery V1 (113 evaluations)Science51.933.9−18.0
SWE-bench VerifiedCode96.778.8−17.9
Artificial Analysis Intelligence Index v4.3.2Indexes56.038.4−17.6
HardGeoBench Country AccuracyMultimodal62.545.0−17.5
GDPval-AA v2.1Professional Work17431357−17.0
SimpleQA Verified (Epoch anti-abstention)General QA46.563.0+16.5
DeepsecBench — Revalidated V1 (233 issues)Code37.521.0−16.5
GDPval Extended (Mercor held-out rubric score)Professional Work70.954.5−16.4
Extended NYT Connections 940 (Quadratic v1)Games80.596.2+15.7
HardGeoBench Avg DistanceMultimodal6891196−15.7
Humanity's Last ExamGeneral QA56.941.4−15.5
Design Arena Overall Code CategoriesCode13491257−15.3
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine29.314.7−14.7
APEX-1 (Mercor, mean score)Professional Work53.968.4+14.5
LMCAGeneral QA68.354.3−14.0
LLM Chess EloGames11951532+13.9
LiveBench 2.0Agents82.568.7−13.8
HardGeoBench <=100miMultimodal53.840.0−13.7
Conceptual Reasoning IndexIndexes77.167.3−13.4
APEX-Agents 1.1Professional Work83.770.3−13.4
SWE-bench MultilingualCode90.977.7−13.2
BullshitBench v2 LeaderboardSafety60.047.0−13.0
Bio Capabilities Index (SecureBio)Indexes175158−12.4
LisanBench Path LengthGames137879845−12.2
ALE-BenchCode24482108−12.0
Roboflow Vision Evals (October 2026, five tasks)Multimodal77.666.0−11.6
ExploitGymCybersecurity26.615.1−11.5
HardGeoBench <=10miMultimodal40.028.8−11.3
Excel Modeling BenchmarkFinance & Accounting75.764.5−11.2
Artificial Analysis Progress Index v1Indexes351320−10.7
HPCT (SecureBio)Science77.266.8−10.4
LAB-Bench ProtocolQA (SecureBio)Science88.377.9−10.4
MineBench — Bradley-Terry RatingGames19981846−9.2
Roboflow Vision Evals (September 2026 protocol)Multimodal83.874.8−9.0
TxBench Preclinical Pharmacology V1 Clean RerunsScience64.355.3−9.0
Mystery Game PuzzlesGames65.056.0−9.0
BrowseComp (Mercor 130-question subset)Search86.677.7−8.9
ScBench-Long V1 Clean RerunsScience30.321.6−8.7
SpatialBench V1 Clean RerunsScience72.864.1−8.7
APEX-Accounting 1.1Finance & Accounting64.556.3−8.2
FrontierMath Tier 4 (v2) - PrivateMath80.572.5−8.0
Terminal-Bench 2.1 Extended (Mercor held-out)Code45.538.0−7.5
AA-AnalystAgent (pass^5)Professional Work57.550.0−7.5
SpeciEval (Pooled English Runs)Safety91.098.2+7.2
SpeciEval (Refusal-Aware Retries, Historical)Safety91.098.2+7.2
SimpleBenchGeneral QA75.969.0−6.9
VCT v2 (SecureBio)Science52.245.3−6.9
WCB (SecureBio)Science66.559.7−6.8
DeepResearch Bench II (Mercor 130-task set)Search51.745.4−6.3
Public Benefits Bench v1.1Professional Work67.260.9−6.3
DeepSWE v1.1Code70.564.4−6.1
CritPt (avg@5)Science31.425.4−6.0
LAB-Bench LitQA2 (SecureBio)Science55.849.8−6.0
ProgramBenchCode6.50.5−6.0
TaxCalcBench TY25Finance & Accounting18.024.0+6.0
AA-Omniscience IndexIndexes32.320.5−5.9
Olam Social Poker EloGames15261521−5.4
GDP.pdfProfessional Work31.426.0−5.4
BuseyBench SVGMultimodal6.56.0−5.3
Vision Arena OverallMultimodal12631283+5.3
MedXpertQA MM Extended (Mercor held-out)Health & Medicine55.660.8+5.2
ACCoRDGeneral QA85.780.5−5.2
GBENCH One-shot Coding GScoreGames67.462.2−5.1
SWE Atlas - Codebase QnACode50.345.2−5.1
Epoch Capabilities Index (ECI)Indexes166160−5.1
SciCode 1.0.1 (Artificial Analysis)Code61.056.1−4.9
AA-LCR v1.1Search79.784.3+4.7
MMMU ProMultimodal85.881.2−4.6
MedScribeHealth & Medicine91.186.9−4.2
Chess Puzzles (task version 2.0)Games59.063.0+4.0
MedXpertQA MM (200-question set)Health & Medicine80.776.7−4.0
MedCodeHealth & Medicine52.949.1−3.8
RedlineBenchLegal40.344.1+3.8
Ruby LLM Benchmarks — Program FixerCode73.877.6+3.8
LisanBench Difficulty-WeightedGames29833316+3.6
FrontierMath Tiers 1-3 (v2) - PrivateMath88.885.3−3.5
AA-LCR Extended (Mercor held-out)Search73.870.5−3.3
ScBench V1 Clean RerunsScience60.357.1−3.2
ChessBench AccuracyGames75.878.8+3.2
Blueprint-Bench 2Multimodal0.40.4−3.1
Terminal-Bench 2.1Code87.384.3−3.0
FrontierMath ErdősMath2.90.0−2.9
NoCoT-Bench NCRI 15.2General QA95.1100+2.8
SpatialBench-Long V1 Clean RerunsScience34.932.1−2.8
LMArena Agent Arena Bash RecoveryAgents15.09.5−2.7
NoCoT-Bench NCKI (kspine v3)General QA107111+2.5
ForecastBench Preliminary Dataset Brier IndexForecasting61.163.3+2.2
DAXBench v2Code86.288.3+2.1
AdvancedIFChat & Writing83.481.3−2.1
MMLU ProGeneral QA91.088.9−2.1
GDPval (Mercor 213-task rubric score)Professional Work86.584.6−1.9
EpiBench V1 Clean RerunsScience30.228.3−1.9
MBCT (SecureBio)Science58.857.1−1.7
LMArena Text Arena EloChat & Writing14721481+1.5
ARC-AGI-1 Semi-PrivateGeneral QA96.595.0−1.5
CharXiv (descriptive and reasoning combined)Multimodal96.194.6−1.5
CharXiv Extended (Mercor held-out)Multimodal84.282.7−1.5
DTBenchGeneral QA97.396.0−1.3
ARC-AGI-2General QA83.985.0+1.1
Harvey's Legal Agent BenchmarkLegal2.93.8+0.8
PubMedQA (SecureBio)Health & Medicine81.380.5−0.8
ARC-AGI-3 (Semi-Private)General QA1.20.4−0.8
LAB-Bench CloningScenarios (SecureBio)Science97.097.3+0.3
SAGEMath51.851.5−0.2
VCT (SecureBio)Science52.552.3−0.2
GPQA-Bio (SecureBio)Science87.687.4−0.2
AIME 2025Math100.0100.00.0
ObviousBenchGeneral QA100.0100.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0

Claude Sonnet 5.5vsGPT-5.5normalised score marginsData Nov 15, 2026Data: Mercor, Artificial Analysis, benchmarks.bio and 39 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…