Atlas

Compare Models

Claude Fable 5 vs Claude Sonnet 5.5

Claude Sonnet 5.5 leads by 3.1 capability points (≈95% interval +0.6 to +5.7): a clear lead over Claude Fable 5 · P(higher) ≈ 99%

Wins 73 of 111 shared benchmarks (36 lost, 2 tied)

Capability vs. relative cost
Claude Fable 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 Fable 5vsClaude Sonnet 5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins36 – 73 · 111 shared, 2 tied

Comparing matching setups only (37 benchmarks excluded) · suites counted once
← Claude Fable 5Claude Sonnet 5.5 →

Claude Fable 5vsClaude Sonnet 5.5wins by categoryData Nov 15, 2026Data: Mercor, Arena, Epoch AI and 41 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 Fable 5.

SWE Atlas Codebase QnA Extended (Mercor held-out)Code0.044.7+44.7
Vending-Bench 2Games453011165+40.2
InferenceBenchCode8.712.1+39.6
Furniture Assembly BenchmarkMultimodal35.875.0+39.2
KernelBench CUDA Normalized (RTX PRO 6000)Code51.687.0+35.4
ActiveVisionMultimodal7.141.2+34.1
Olam Social Poker EloGames15531526−29.3
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.845.5+28.7
Arena Image-to-WebDevCode16211744+26.9
ChartographyMultimodal34.861.6+26.8
GeneBench-Pro Public Release SubsetScience14.039.0+25.0
SWE-bench Verified Extended (Mercor held-out)Code78.757.3−21.4
Bug Hunt BenchCode27.648.9+21.3
Terminal-Bench 4.0 (Artificial Analysis)Code42.463.6+21.2
KernelBench Mega Normalized (RTX PRO 6000)Code69.490.6+21.2
LMArena Code Arena WebDevCode16251773+19.3
AdvancedIFChat & Writing65.283.4+18.2
ChessBench CoherenceGames82.664.6−18.0
Terminal-Bench 4.0Code44.561.8+17.3
SWE Atlas - Codebase QnACode33.150.3+17.2
LisanBench Difficulty-WeightedGames45622983−16.9
LMArena Agent Arena Net Improvement (Steering Burden)Agents7.316.0+16.3
GameBench 2Games75.291.4+16.2
AA-Briefcase v1.1 EloAgents15361816+15.4
DAYJOB: FinanceFinance & Accounting9.024.0+15.0
NoCoT-Bench NCKI (kspine v3)General QA128107−14.6
NoCoT-Bench NCRI 15.2General QA11995.1−13.2
MMMU ProMultimodal72.785.8+13.1
Concept Synth INDUCTION Challenge100 Round 1General QA31.044.0+13.0
Mystery Game PuzzlesGames52.065.0+13.0
Extended NYT Connections 940 (Quadratic v1)Games92.780.5−12.2
APEX-1 (Mercor, mean score)Professional Work66.053.9−12.1
Chess Puzzles (task version 2.0)Games47.059.0+12.0
TaxCalcBench TY25Finance & Accounting30.018.0−12.0
Vision Arena OverallMultimodal13091263−12.0
Physical-Engineering's Last Exam (hidden test)Professional Work33.345.1+11.8
Humanity's Last ExamGeneral QA57.846.1−11.7
MMLU ProGeneral QA79.491.0+11.6
PLE (Physical-Engineering's Last Exam)Professional Work33.845.2+11.4
ChessBench EloGames1046857−11.2
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.475.0+10.6
FrontierMath Tier 4 (v2) - PrivateMath90.280.5−9.7
DAYJOB: HealthcareHealth & Medicine5.615.2+9.6
APEX-Agents 1.1Professional Work74.583.7+9.2
Eval Connections One-Shot with TrapsGames85.076.0−9.0
AA-AnalystAgent (pass^5)Professional Work48.857.5+8.7
YC-BenchGames19775731741315−8.2
LMArena Agent Arena Praise vs ComplaintAgents14.330.5+8.1
MineBench — Bradley-Terry RatingGames18691998+7.8
Concept Synth INDUCTION Challenge64 Round 1General QA42.250.0+7.8
Humanity's Last Exam DiamondGeneral QA45.653.3+7.7
APEX-SWECode58.966.4+7.5
LMArena Agent Arena Confirmed SuccessAgents5.119.5+7.2
LMCAGeneral QA61.168.3+7.1
CharXiv (descriptive and reasoning combined)Multimodal89.196.1+7.0
SWE-TogetherCode68.862.4−6.4
Conceptual Reasoning IndexIndexes72.777.1+6.1
Roboflow Vision Evals (October 2026, five tasks)Multimodal71.777.6+5.9
GDPval Extended (Mercor held-out rubric score)Professional Work65.370.9+5.6
LMArena Text Arena EloChat & Writing15041472−5.5
ARC-AGI-2General QA89.283.9−5.3
Roboflow Vision Evals (September 2026 protocol)Multimodal78.783.8+5.1
ProofBench v1.1Math95.0100.0+5.0
MedXpertQA MM (200-question set)Health & Medicine76.280.7+4.5
WeirdML v3Science15.619.9+4.4
BrowseComp (Mercor 130-question subset)Search82.586.6+4.1
BullshitBench v2 LeaderboardSafety56.060.0+4.0
SWE-bench MultilingualCode94.890.9−3.9
APEX-Accounting 1.1Finance & Accounting61.064.5+3.5
CharXiv Extended (Mercor held-out)Multimodal80.784.2+3.5
HardGeoBench Avg DistanceMultimodal800689+3.4
Ruby LLM Benchmarks — Program FixerCode77.273.8−3.4
MaxxBenchCode63.566.9+3.4
LMArena Agent Arena Steering BurdenAgents7.914.5+3.3
MedCodeHealth & Medicine56.152.9−3.1
FrontierMath ErdősMath0.02.9+2.9
LMArena Agent Arena Bash RecoveryAgents9.615.0+2.7
AA-LCR Extended (Mercor held-out)Search71.173.8+2.7
GDPval (Mercor 213-task rubric score)Professional Work83.886.5+2.7
HardGeoBench <=100miMultimodal56.353.8−2.5
RuneBenchGames5.75.5−2.4
RedlineBenchLegal42.440.3−2.1
Excel Modeling BenchmarkFinance & Accounting73.775.7+2.0
Vibe Code Bench v1.1Code90.492.4+2.0
ARC-AGI-1 Semi-PrivateGeneral QA98.596.5−2.0
GBENCH One-shot Coding GScoreGames65.667.4+1.8
FrontierMath Tiers 1-3 (v2) - PrivateMath87.088.8+1.8
GDP.pdfProfessional Work29.831.4+1.6
ChessBench AccuracyGames77.175.8−1.4
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.955.6−1.3
HardGeoBench <=10miMultimodal41.340.0−1.2
HardGeoBench Country AccuracyMultimodal61.362.5+1.2
SWE-bench VerifiedCode95.596.7+1.2
LMArena Agent Arena Bash ReliabilityAgents10.412.7+1.2
LiveBench 2.0Agents83.682.5−1.2
DTBenchGeneral QA98.497.3−1.1
FrontierCode 1.1 Extended - Pass RateCode70.969.9−1.0
ACCoRDGeneral QA85.085.7+0.7
Blueprint-Bench 2Multimodal0.40.4+0.7
Terminal-Bench 2.1Code88.087.3−0.7
ObviousBenchGeneral QA99.3100.0+0.7
ALE-BenchCode24632448−0.5
FrontierCode 1.1 Extended - ScoreCode64.964.4−0.5
EnterpriseBench: CoreCraftProfessional Work70.370.8+0.5
Harvey LAB-AA v1.0 (criterion pass rate)Legal93.693.1−0.5
Prophet Arena Arena Score (Fixed Context)Forecasting84.985.3+0.4
LisanBench Path LengthGames1368013787+0.3
LMArena Agent Arena Tool HallucinationAgents0.40.9+0.2
SAGEMath51.951.8−0.1
AIME 2025Math100.0100.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0

Claude Fable 5vsClaude Sonnet 5.5normalised score marginsData Nov 15, 2026Data: Mercor, Arena, Epoch AI 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…