Atlas

Compare Models

Claude Haiku 5.5 vs Claude Sonnet 5.5

Claude Sonnet 5.5 leads by 5.3 capability points (≈95% interval +2.9 to +7.7): a clear lead over Claude Haiku 5.5 · P(higher) ≈ 100%

Wins 177 of 195 shared benchmarks (12 lost, 6 tied)

Capability vs. relative cost
Claude Haiku 5.5Claude Sonnet 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 Haiku 5.5vsClaude Sonnet 5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins12 – 177 · 195 shared, 6 tied

Comparing matching setups only (20 benchmarks excluded) · suites counted once
← Claude Haiku 5.5Claude Sonnet 5.5 →

Claude Haiku 5.5vsClaude Sonnet 5.5wins by categoryData Nov 15, 2026Data: Artificial Analysis, Mercor, Vals AI and 53 othersVillage LLM Atlasatlas.village.dev

Margins

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

InferenceBenchCode4.412.1+91.1
KernelBench Mega Normalized (RTX PRO 6000)Code0.090.6+90.6
ARC-AGI-3 (Semi-Private)General QA2.872.4+69.6
Vending-Bench 2Games360211165+45.8
KernelBench CUDA Normalized (RTX PRO 6000)Code41.487.0+45.6
AA-PersonalityBench Organization indexChat & Writing41.084.8+44.2
Toloka Arena (v3)Agents21.064.0+43.0
Terminal-Bench Science 0.1 (Vals)Science10.047.1+37.1
AutomationBench-AA 1.0.6Agents36.071.8+35.8
IOI (Vals v2)Code47.383.1+35.8
Mystery Game PuzzlesGames30.065.0+35.0
Code Migration v1.1Code52.486.9+34.5
FrontierMath Tier 4 (v2) - PrivateMath46.380.5+34.1
Terminal-Bench-Science 0.1Science20.053.3+33.3
Artificial Analysis Coding Agent Index v1.5Indexes36.868.4+31.5
Terminal-Bench 4.0 (Artificial Analysis)Code32.863.6+30.8
BrowseComp (Mercor 130-question subset)Search56.386.6+30.3
AA-3D v1.0 EloCode9981204+29.8
LMArena Agent Arena Net Improvement (Steering Burden)Agents0.516.0+29.2
Terminal-Bench 4.0Code33.361.8+28.5
Bug Hunt BenchCode20.548.9+28.4
Furniture Assembly BenchmarkMultimodal47.575.0+27.5
Decision-Bench v1.1Professional Work25.351.2+25.9
Code MigrationCode44.669.8+25.3
MineBench — Bradley-Terry RatingGames15821998+25.2
Humanity's Last Exam DiamondGeneral QA29.653.3+23.7
LMArena Code Arena WebDevCode15941773+23.4
AA-PersonalityBench Transparency indexChat & Writing77.2100+23.1
SimpleQA Verified (Epoch anti-abstention)General QA23.846.5+22.7
ARC-AGI-2General QA61.383.9+22.6
AA-PersonalityBench Imagination indexChat & Writing63.884.8+21.2
Code2Video-AA EloCode11921443+20.4
BuseyBench SVGMultimodal4.66.5+19.7
HardGeoBench Avg DistanceMultimodal1319683+19.6
Artificial Analysis Strategy & Ops IndexIndexes39.158.1+18.9
FrontierSWE V2Code43.861.9+18.1
Artificial Analysis Intelligence Index v5.0.1Indexes36.554.5+18.1
Artificial Analysis Intelligence Index v5.0Indexes36.254.2+18.0
DeepResearch Bench II (Mercor 130-task set)Search34.351.7+17.4
AA-PersonalityBench Slop indexChat & Writing10.527.7+17.3
LMCAGeneral QA51.168.3+17.2
ExploitGymCybersecurity9.426.6+17.1
LMArena Agent Arena Praise vs ComplaintAgents-3.530.5+17.0
Artificial Analysis Healthcare & Medical IndexIndexes37.554.2+16.7
Excel Modeling BenchmarkFinance & Accounting59.075.7+16.7
GameBench 2Games75.191.4+16.4
Conceptual Reasoning IndexIndexes65.377.1+16.2
ChessBench CoherenceGames80.764.6−16.1
ALE-BenchCode19962448+16.0
Cua Driver Bench v2 (Linux, foreground)Agents68.084.0+16.0
TaxCalcBench TY25Finance & Accounting2.018.0+16.0
Artificial Analysis Engineering IndexIndexes42.558.5+16.0
FrontierML ($200 wallet)Agents34.950.4+15.5
APEX-Agents 1.1Professional Work68.383.7+15.4
ChartographyMultimodal46.461.6+15.2
Extended NYT Connections 940 (Quadratic v1)Games65.780.5+14.8
NoCoT-Bench NCKI (kspine v3)General QA85.9107+14.7
NoCoT-Bench NCRI 15.2General QA68.395.1+14.7
DeepFinanceFinance & Accounting40.054.6+14.6
FrontierCode 2.0 - Pass RateCode31.946.4+14.5
SWE-bench Multimodal (Vals v1, 480 Tasks)Code44.058.3+14.4
MysteryMechanismScience35.149.1+14.0
FrontierMath Tiers 1-3 (v2) - PrivateMath75.188.8+13.7
Artificial Analysis Legal IndexIndexes38.752.2+13.5
AA-Briefcase v1.1 EloAgents15731816+13.4
FrontierCode 2.0 - ScoreCode28.942.2+13.3
MedXpertQA MM (200-question set)Health & Medicine67.880.7+12.9
Artificial Analysis Intelligence Index v4.3.2Indexes43.456.0+12.6
Cua Driver Bench v2 (macOS, background)Agents45.057.5+12.5
Humanity's Last Exam (Text-Only)General QA42.755.0+12.3
Artificial Analysis Finance & Accounting IndexIndexes43.655.8+12.2
GBENCH One-shot Coding GScoreGames55.667.4+11.7
Medical Long Context Reasoning (MLCR-AA)Health & Medicine63.375.0+11.7
OSWorld 2.1 offline - reported partial reward, step limit unspecifiedAgents72.483.9+11.5
Design Arena Overall Code CategoriesCode12801349+11.4
Artificial Analysis Economics IndexIndexes42.754.1+11.4
APEX-SWECode55.166.4+11.3
SM BenchSafety74.385.4+11.1
MKA (Artificial Analysis)Professional Work41.052.0+11.0
ProofBench v1.1Math89.0100.0+11.0
τ³-Banking (Mercor run, bm25_grep retrieval)Agents43.054.0+11.0
AA-Omniscience IndexIndexes10.732.3+10.8
APEX-1 (Mercor, mean score)Professional Work64.753.9−10.8
DeepSWE v1.1Code59.970.5+10.6
VIALS-AA (pass^3)Multimodal49.159.6+10.6
Excel Modeling Benchmark v1.1Finance & Accounting65.575.9+10.4
AA-LCR Extended (Mercor held-out)Search63.873.8+10.0
Roboflow Vision Evals — Counting, Judged (September)Multimodal73.082.9+9.9
AA-PersonalityBench Informativeness indexChat & Writing75.384.8+9.6
AA-PersonalityBench Conviction indexChat & Writing67.658.1−9.6
ChessBench AccuracyGames66.975.8+9.5
GDPval-AA v2.1Professional Work16291838+9.2
Roboflow Vision Evals — Reasoning, Judged (September)Multimodal74.883.9+9.1
Eval Connections One-Shot with TrapsGames67.076.0+9.0
LMArena Agent Arena Confirmed SuccessAgents1.619.5+9.0
CritPt v2 (agentic, avg@5)Science66.375.1+8.9
CritPt (avg@5)Science22.631.4+8.9
Roboflow Vision Evals — Detection mAP@75 (September)Multimodal54.363.1+8.8
LMArena Text Arena EloChat & Writing14221472+8.7
Roboflow Vision Evals — Detection mAP@50 (September)Multimodal68.276.8+8.6
Artificial Analysis Progress Index v1Indexes326351+8.6
SWE-Bench Pro V2 — FullCode87.295.8+8.6
WeirdML v3Science11.719.9+8.2
SWE Atlas v1.1 - OverallCode59.467.5+8.1
Roboflow Vision Evals (October 2026, five tasks)Multimodal69.677.6+8.0
SWE-bench VerifiedCode88.996.7+7.8
Terminal-Bench 2.1 Extended (Mercor held-out)Code37.745.5+7.8
HardGeoBench Country AccuracyMultimodal51.358.8+7.5
Cua Driver Bench v2 (Windows, background)Agents20.027.5+7.5
ARC-AGI-1 Semi-PrivateGeneral QA89.096.5+7.5
MedXpertQA MM Extended (Mercor held-out)Health & Medicine48.455.6+7.2
Roboflow Vision Evals — Counting, Strict (September)Multimodal73.080.2+7.2
Immigration BenchLegal61.869.0+7.2
CursorBench 4.0Code48.455.5+7.1
Humanity's Last Exam (with tools)General QA57.464.5+7.1
LiveBench 2.0Agents75.682.5+6.8
MMMU ProMultimodal79.185.8+6.7
Roboflow Vision Evals — Detection mAP@50:95 (September)Multimodal53.059.5+6.5
FrontierCode 1.1 Extended - ScoreCode58.164.4+6.3
Blueprint-Bench 2Multimodal0.30.4+6.3
HardGeoBench <=10miMultimodal30.036.3+6.3
HLAB-Vals-1.1Legal4.210.4+6.3
LMArena Agent Arena Steering BurdenAgents2.414.5+6.1
HealthBench ProfessionalHealth & Medicine64.870.8+6.0
SciCode 1.0.1 (Artificial Analysis)Code55.061.0+6.0
Agent Initiative BenchmarkAgents6.06.7+6.0
SAGEMath45.851.8+6.0
SWE-bench MultilingualCode85.190.9+5.8
FrontierCode 1.1 Extended - Pass RateCode64.169.9+5.7
Roboflow Vision Evals — Identification, Judged (September)Multimodal86.591.7+5.2
GDP.pdf (Artificial Analysis, All-pass)Professional Work20.825.8+5.0
HardGeoBench <=100miMultimodal45.050.0+5.0
VIALS-AA (average of 3)Multimodal60.565.4+5.0
Epoch Capabilities Index (ECI)Indexes160166+4.9
GDPval Extended (Mercor held-out rubric score)Professional Work66.070.9+4.9
HealthBenchHealth & Medicine61.265.9+4.6
DAXBench v2Code81.986.2+4.3
ACCoRDGeneral QA81.585.7+4.1
SpeciEval (Pooled English Runs)Safety86.991.0+4.0
SWE-bench Verified Extended (Mercor held-out)Code53.357.3+4.0
DualEntry Accounting AI BenchmarkFinance & Accounting76.280.2+4.0
ProgramBenchCode2.56.5+4.0
Chess Puzzles (task version 2.0)Games55.059.0+4.0
Public Benefits Bench v1.1Professional Work63.267.2+4.0
AA-PersonalityBench Expressiveness indexChat & Writing77.273.3−3.8
AA-PersonalityBench Warmth indexChat & Writing69.573.3+3.8
AA-AnalystAgent (pass^5)Professional Work53.857.5+3.7
CVDP (517-problem set)Code57.060.6+3.6
MMLU ProGeneral QA87.491.0+3.6
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine58.562.1+3.6
GDPxProfessional Work4.98.2+3.3
o11y-bench 1.1 Pass^3Agents87.590.6+3.1
BullshitBench v2 LeaderboardSafety57.060.0+3.0
BioSimBench 1.0Science13.016.0+3.0
LisanBench Difficulty-WeightedGames27112983+2.9
DistressBenchSafety71.774.6+2.9
LisanBench Path LengthGames1289713787+2.8
RCM-Bench-1Health & Medicine88.491.1+2.7
LMArena Agent Arena Bash ReliabilityAgents7.412.7+2.7
AdvancedIFChat & Writing80.883.4+2.6
Hermes BenchAgents71.073.5+2.5
DTBenchGeneral QA95.097.3+2.3
APEX-Accounting 1.1Finance & Accounting62.264.5+2.3
Terminal-Bench 2.1Code85.087.3+2.3
LLM Chess EloGames11431195+2.1
CharXiv (descriptive and reasoning combined)Multimodal94.196.1+2.0
ChessBenchGames41.339.3−2.0
Cua Driver Bench v2 (macOS, foreground)Agents76.078.0+2.0
Vibe Code Bench v1.1Code90.492.4+1.9
SAGE v1.1Math89.888.0−1.8
Harvey's Legal Agent BenchmarkLegal1.32.9+1.7
SWE Atlas - Codebase QnACode48.750.3+1.6
MedScribeHealth & Medicine89.591.1+1.6
LMArena Agent Arena Tool HallucinationAgents-2.10.9+1.5
MedCodeHealth & Medicine51.552.9+1.4
AA-LCR v1.1Search78.379.7+1.3
CharXiv Extended (Mercor held-out)Multimodal82.984.2+1.3
SWE-Marathon v1.1Code50.051.3+1.2
OTIS Mock AIME 2024-2025Math98.9100.0+1.1
Roboflow Vision Evals — Identification, Strict (September)Multimodal80.281.3+1.1
ChessBench EloGames875857−1.1
GDPval (Mercor 213-task rubric score)Professional Work85.586.5+1.0
VIALS-AA (pass@3)Multimodal71.470.8−0.6
ForecastBench Preliminary Dataset Brier IndexForecasting60.861.1+0.3
RedlineBenchLegal40.640.3−0.3
Prophet Arena Arena Score (Agentic, OpenCode)Forecasting85.985.7−0.2
Ramp Accounting BenchFinance & Accounting42.342.5+0.2
VulcanBench Frontier v4Code92.292.0−0.1
Prophet Arena Arena Score (Fixed Context)Forecasting85.485.3−0.1
AIME 2025Math100.0100.00.0
ClaimBenchFinance & Accounting100.0100.00.0
Cua Driver Bench v2 (Linux, background)Agents12.512.50.0
Cua Driver Bench v2 (Windows, foreground)Agents70.070.00.0
MazeBench ASCIIGames1.01.00.0
Roboflow Vision Evals — Reasoning, Strict (September)Multimodal72.872.80.0

Claude Haiku 5.5vsClaude Sonnet 5.5normalised score marginsData Nov 15, 2026Data: Artificial Analysis, Mercor, Vals AI and 53 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…