Atlas

Compare Models

Claude Fable 5.1 vs Claude Sonnet 5.5

Claude Sonnet 5.5 leads by 0.9 capability points (≈95% interval −1.7 to +3.5): statistically tied · P(higher) ≈ 75%

Wins 98 of 183 shared benchmarks (79 lost, 6 tied)

Capability vs. relative cost
Claude Fable 5.1Claude 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 5.1vsClaude Sonnet 5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins79 – 98 · 183 shared, 6 tied

Comparing matching setups only (15 benchmarks excluded) · suites counted once
← Claude Fable 5.1Claude Sonnet 5.5 →

Claude Fable 5.1vsClaude Sonnet 5.5wins by categoryData Nov 15, 2026Data: Mercor, Artificial Analysis, Surge AI and 62 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 5.1.

SPU (Science Prediction Universe)Science-0.50.3+44.2
AA-PersonalityBench Conviction indexChat & Writing10058.1−42.3
KernelBench CUDA Normalized (RTX PRO 6000)Code51.687.0+35.4
Olam Social Poker EloGames14941526+34.8
InferenceBenchCode9.312.1+32.8
Terminal-Bench 2.1 Extended (Mercor held-out)Code19.545.5+26.0
ActiveVisionMultimodal15.341.2+25.9
KernelBench Mega Normalized (RTX PRO 6000)Code64.790.6+25.9
AA-PersonalityBench Expressiveness indexChat & Writing98.173.3−25.0
AA-PersonalityBench Organization indexChat & Writing60.084.8+25.0
SimpleQA Verified (Epoch anti-abstention)General QA70.846.5−24.3
TaxCalcBench TY25Finance & Accounting42.018.0−24.0
AA-PersonalityBench Warmth indexChat & Writing94.373.3−21.2
HoneyBench v0.2 (hack rate)Safety61.841.6+20.2
APEX-1 (Mercor, mean score)Professional Work73.853.9−19.9
NoCoT-Bench NCRI 15.2General QA12895.1−18.1
Android Bench 2.0Code22.740.0+17.3
LisanBench Difficulty-WeightedGames45732983−17.0
BullshitBench v2 LeaderboardSafety77.060.0−17.0
Concept Synth INDUCTION Challenge64 Round 1General QA34.450.0+15.6
BioSecBench-Structure V1 (58 evaluations)Science6.321.8+15.5
ChartographyMultimodal46.261.6+15.4
NoCoT-Bench NCKI (kspine v3)General QA128107−14.7
ALE-BenchCode28492448−14.2
ChessBench EloGames1095857−14.2
Humanity's Last Exam DiamondGeneral QA51.337.5−13.8
BioSecBench-RespiratoryVirus-Surveillance V1 (110 evaluations)Science3.617.3+13.6
GeneBench-Pro Public Release SubsetScience26.039.0+13.0
AA-PersonalityBench Informativeness indexChat & Writing96.284.8−11.5
Visa Agentic Commerce BenchmarkAgents75.664.4−11.1
Chess Puzzles (task version 2.0)Games48.059.0+11.0
Concept Synth INDUCTION Challenge100 Round 1General QA33.044.0+11.0
SimpleBenchGeneral QA86.675.9−10.7
sudo L7Code31.141.7+10.6
MedXpertQA MM Extended (Mercor held-out)Health & Medicine66.055.6−10.4
FrontierCode 2.0 - Pass RateCode36.446.4+10.0
OpenWorldEngineer (Mean@8)Code29.139.0+9.9
Code2Video-AA EloCode13251443+9.6
RuneBenchGames6.15.5−9.5
Extended NYT Connections 940 (Quadratic v1)Games90.080.5−9.5
AA-3D v1.0 EloCode11391204+9.4
DAYJOB: Data ScienceProfessional Work26.035.3+9.3
FrontierCode 2.0 - ScoreCode33.042.2+9.3
DeepFinanceFinance & Accounting45.654.6+9.0
TxBench-Antibody-Development V1 (60 evaluations)Science27.836.7+8.9
HardGeoBench <=10miMultimodal45.036.3−8.7
FrontierML ($200 wallet)Agents41.950.4+8.5
Agent Initiative BenchmarkAgents7.56.7−8.5
BrowseComp (Mercor 130-question subset)Search78.386.6+8.3
DAYJOB: AccountingFinance & Accounting27.235.2+8.0
AA-Briefcase v1.1 EloAgents16711816+8.0
Bug Hunt BenchCode41.048.9+7.9
AA-LCR v1.1Search87.579.7−7.8
FrontierMath Tier 4 (v2) - PrivateMath87.880.5−7.3
Ramp Accounting BenchFinance & Accounting49.742.5−7.2
OSWorld 2.1 offline - reported partial reward, step limit unspecifiedAgents76.883.9+7.1
Mystery Game PuzzlesGames58.065.0+7.0
SWE Atlas - Codebase QnACode43.350.3+7.0
Toloka Arena (v3)Agents71.064.0−7.0
SWE-TogetherCode69.362.4−6.9
Vision Arena OverallMultimodal12891263−6.8
ComplexConstraints (November 2026 rescoring)Chat & Writing60.467.2+6.8
SWE Atlas Codebase QnA Extended (Mercor held-out)Code38.044.7+6.7
APEX-Agents 1.1Professional Work77.183.7+6.6
EnterpriseBench: CoreCraftProfessional Work77.470.8−6.6
SWE Atlas v1.1 - OverallCode74.167.5−6.6
Immigration BenchLegal75.469.0−6.5
Artificial Analysis Coding Agent Index v1.5Indexes62.268.4+6.1
ARC-AGI-2General QA90.083.9−6.1
WeirdML v3Science26.019.9−6.0
GDP.xlsxProfessional Work23.129.1+6.0
ChessBenchGames45.339.3−6.0
Eval Connections One-Shot with TrapsGames82.076.0−6.0
WellnessBenchSafety51.057.0+6.0
Terminal-Bench 2.1Code81.687.3+5.7
SWE-Marathon v1.1Code45.651.3+5.6
Real-SWE 1.1Code45.651.2+5.6
DAYJOB: HealthcareHealth & Medicine9.615.2+5.6
RiemannBench (November 2026 rescoring)Math71.276.8+5.6
LiveBench 2.0Agents88.082.5−5.5
DAXBench v2Code91.586.2−5.3
LMArena Text Arena EloChat & Writing15031472−5.3
BuseyBench SVGMultimodal7.06.5−5.0
Furniture Assembly BenchmarkMultimodal70.075.0+5.0
HardGeoBench Avg DistanceMultimodal845683+5.0
DualEntry Accounting AI BenchmarkFinance & Accounting85.180.2−4.9
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine57.262.1+4.8
ChessBench AccuracyGames71.375.8+4.8
Arena Image-to-WebDevCode17241744+4.5
MineBench — Bradley-Terry RatingGames19281998+4.2
DAYJOB: FinanceFinance & Accounting19.824.0+4.2
LisanBench Path LengthGames1510013787−4.1
Humanity's Last ExamGeneral QA60.956.9−4.0
SWE-bench Verified Extended (Mercor held-out)Code61.357.3−4.0
LMArena Code Arena WebDevCode17421773+4.0
Terminal-Bench 4.0Code57.961.8+3.9
Medical Long Context Reasoning (MLCR-AA)Health & Medicine71.175.0+3.9
AA-PersonalityBench Slop indexChat & Writing23.827.7+3.8
SpeciEval (Pooled English Runs)Safety87.291.0+3.8
SpeciEval (Refusal-Aware Retries, Historical)Safety87.291.0+3.8
CursorBench 4.0Code51.855.5+3.7
Harvey LAB-AA v1.1 (hallucination-gated)Legal6.42.8−3.6
SM BenchSafety89.085.4−3.6
DeepResearch Bench II (Mercor 130-task set)Search55.051.7−3.3
SAGEMath48.551.8+3.2
GDPval-AA v2.1Professional Work17651838+3.2
DeepSWE v1.1Code67.370.5+3.2
FrontierCode 1.1 Extended - ScoreCode61.464.4+3.1
DistressBenchSafety71.574.6+3.0
MedXpertQA MM (200-question set)Health & Medicine83.780.7−3.0
Design Arena Overall Code CategoriesCode13311349+3.0
FrontierCode 1.1 Extended - Pass RateCode66.969.9+3.0
Conceptual Reasoning IndexIndexes74.977.1+2.9
FrontierMath ErdősMath0.02.9+2.9
Roboflow Vision Evals (October 2026, five tasks)Multimodal74.777.6+2.9
LMCAGeneral QA65.568.3+2.8
APEX-SWECode63.666.4+2.8
ChessBench CoherenceGames61.964.6+2.7
GameBench 2Games94.091.4−2.6
Blueprint-Bench 2Multimodal0.40.4−2.6
LMArena Agent Arena Confirmed SuccessAgents14.419.5+2.5
Roboflow Vision Evals (September 2026 protocol)Multimodal81.383.8+2.5
APEX-Accounting 1.1Finance & Accounting66.764.5−2.2
GDPval Extended (Mercor held-out rubric score)Professional Work68.770.9+2.2
Ruby LLM Benchmarks — Program FixerCode76.073.8−2.2
LMArena Agent Arena Net Improvement (Steering Burden)Agents14.816.0+2.1
SciCode 1.0.1 (Artificial Analysis)Code63.161.0−2.1
CharXiv Extended (Mercor held-out)Multimodal82.284.2+2.0
CVDP (517-problem set)Code62.660.6−2.0
MMLU ProGeneral QA93.091.0−2.0
ACCoRDGeneral QA83.785.7+2.0
HealthBenchHealth & Medicine63.965.9+2.0
Clipping Bench v1.0Games1068886−2.0
GBENCH One-shot Coding GScoreGames69.367.4−1.9
AA-PersonalityBench Imagination indexChat & Writing86.784.8−1.9
AA-PersonalityBench Transparency indexChat & Writing98.1100+1.9
LMArena Agent Arena Bash RecoveryAgents11.615.0+1.7
HealthBench ProfessionalHealth & Medicine69.270.8+1.6
CritPt v2 (agentic, avg@5)Science76.675.1−1.4
FrontierMath Tiers 1-3 (v2) - PrivateMath90.288.8−1.4
micro1 PharmaBenchHealth & Medicine12.811.4−1.4
MysteryMechanismScience47.749.1+1.4
LMArena Agent Arena Bash ReliabilityAgents10.112.7+1.3
HardGeoBench <=100miMultimodal52.553.8+1.3
HardGeoBench Country AccuracyMultimodal61.362.5+1.2
DAYJOB: TaxFinance & Accounting9.610.8+1.2
ARC-AGI-1 Semi-PrivateGeneral QA97.596.5−1.0
CharXiv (descriptive and reasoning combined)Multimodal95.196.1+1.0
MazeBench ASCIIGames2.01.0−1.0
Excel Modeling BenchmarkFinance & Accounting76.775.7−1.0
Excel Modeling Benchmark v1.1Finance & Accounting76.775.9−0.8
AA-LCR Extended (Mercor held-out)Search73.073.8+0.8
AdvancedIFChat & Writing84.283.4−0.8
Financial Coaching BenchFinance & Accounting75.774.9−0.8
ObviousBenchGeneral QA99.3100.0+0.7
SpireClimbGames2.01.3−0.7
OSWorld 2.1 full - reported partial reward, step limit unspecifiedAgents80.780.1−0.6
HANDBOOK.md Agents v1.1Professional Work47.147.7+0.6
MedCodeHealth & Medicine53.552.9−0.6
Epoch Capabilities Index (ECI)Indexes167166−0.6
LMArena Agent Arena Praise vs ComplaintAgents31.630.5−0.6
MissionReportProfessional Work63.363.8+0.5
DAYJOB: LawLegal26.527.0+0.5
Qodo AI Code Review Benchmark (125 PRs)Code53.554.0+0.5
Prophet Arena Arena Score (Agentic, OpenCode)Forecasting86.085.7−0.3
GDPxProfessional Work8.58.2−0.3
MMMU ProMultimodal86.185.8−0.3
CritPt (avg@5)Science31.131.4+0.3
DTBenchGeneral QA97.697.3−0.3
Prophet Arena Arena Score (Fixed Context)Forecasting85.085.3+0.3
Harvey LAB-AA v1.0 (criterion pass rate)Legal93.393.1−0.2
SWE-bench MultilingualCode91.190.9−0.2
MedScribeHealth & Medicine91.391.1−0.2
LMArena Agent Arena Steering BurdenAgents14.814.5−0.1
SWE-bench VerifiedCode96.696.7+0.1
GDPval (Mercor 213-task rubric score)Professional Work86.686.5−0.1
LMArena Agent Arena Tool HallucinationAgents0.80.90.0
AA-AnalystAgent (pass^5)Professional Work57.557.50.0
AIME 2025Math100.0100.00.0
Dual-Arm Sim (task completion)Agents0.00.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
ProofBench v1.1Math100.0100.00.0
τ³-Banking (Mercor run, bm25_grep retrieval)Agents54.054.00.0

Claude Fable 5.1vsClaude Sonnet 5.5normalised score marginsData Nov 15, 2026Data: Mercor, Artificial Analysis, Surge AI and 62 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…