Atlas

Compare Models

Claude Fable 5 vs GPT-5.5

Claude Fable 5 leads by 3.5 capability points (≈95% interval +1.2 to +5.7): a clear lead over GPT-5.5 · P(higher) ≈ 100%

Wins 154 of 207 shared benchmarks (47 lost, 6 tied)

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

Benchmark wins154 – 47 · 207 shared, 6 tied

Comparing matching setups only (194 benchmarks excluded) · suites counted once
← Claude Fable 5GPT-5.5 →

Claude Fable 5vsGPT-5.5wins by categoryData Nov 15, 2026Data: Mercor, Vals AI, Lech Mazur Benchmarks and 66 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 Fable 5.

KernelBench Hard Normalized (H100)Code99.20.0−99.2
VCT-Refusal (CAIS)Safety100.016.0−84.0
HPCT Refusal (CAIS)Safety100.029.0−71.0
KernelBench Mega Normalized (H100)Code78.723.1−55.6
MirrorCode (ML, +Private, 2L)Code63.910.0−53.9
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.417.8−46.7
SWE-bench Verified Extended (Mercor held-out)Code78.739.3−39.4
GSO Opt@1Code78.440.2−38.2
Olam Social Poker EloGames15531521−34.8
Antidote: Everyday EditionSafety10991013−30.7
ExploitBench - release-reported exploit progressCybersecurity78.047.9−30.1
LLM Debate BenchmarkChat & Writing17271528−28.3
Terminal-Bench 4.0 (Artificial Analysis)Code42.414.6−27.8
ProofBenchMath77.050.0−27.0
YC-BenchGames19775731206045−26.9
BioTIER RefusalSafety100.073.4−26.6
BrowseComp Extended (Mercor held-out)Search44.519.0−25.5
LLM Position Bias First-Shown Pick RateChat & Writing58.969.9−22.0
LLM Position Bias First-Option LiftChat & Writing8.919.9−22.0
SWE-Bench Pro (Public Dataset)Code80.358.6−21.7
AA-Briefcase EloAgents15301158−21.6
LLM Position Bias Decisive Pair CoverageChat & Writing72.093.3+21.3
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.838.0+21.2
Hemingway-benchChat & Writing11101035−20.8
Vibe Code Bench v1.1Code90.469.8−20.5
Drone-BenchAgents84.363.9−20.4
EnterpriseBench: CoreCraftProfessional Work70.351.3−19.0
PointerbenchMultimodal78.759.9−18.8
OSWorld 2.0 500-step Partial ScoreAgents66.147.5−18.6
EBR-bench v2Games39.521.0−18.6
PrinzBenchLegal56.074.0+18.2
Vending-Bench 2Games45307524+18.1
ChessBench EloGames10461349+18.0
GlobeBenchGeneral QA7.325.2+17.9
FrontierMath Tier 4 (v2) - PrivateMath90.272.5−17.7
Humanity's Last Exam DiamondGeneral QA45.627.9−17.7
ChessBench CoherenceGames82.6100.0+17.4
SWE Atlas Codebase QnA Extended (Mercor held-out)Code0.017.3+17.3
GDPval-AA v2Professional Work17231372−17.3
LLM Position Bias Order Flip RateChat & Writing27.344.4−17.1
SWE-bench MultilingualCode94.877.7−17.1
ChessBench — Historical Output RepairGames46.029.3−16.7
HiL-Bench (Human-in-Loop Benchmark)Agents56.339.7−16.7
HardGeoBench Country AccuracyMultimodal61.345.0−16.3
HardGeoBench <=100miMultimodal56.340.0−16.2
AdvancedIFChat & Writing65.281.3+16.1
Chess Puzzles (task version 2.0)Games47.063.0+16.0
GameBench 2Games75.259.9−15.3
BrokenArXiv — OverallMath48.863.9+15.0
LiveBench 2.0Agents83.668.7−14.9
HANDBOOK.mdProfessional Work36.221.5−14.7
Public Benefits Bench v1 - BothProfessional Work71.757.2−14.4
Pencil Puzzle Bench - AgenticGames97.683.3−14.3
Humanity's Last ExamGeneral QA57.843.8−14.0
APEX-SWECode54.840.8−14.0
Public Benefits Bench v1 - Web SearchProfessional Work67.753.9−13.9
Public Benefits Bench v1 - NeitherProfessional Work63.449.7−13.7
LisanBench Difficulty-WeightedGames45623316−13.3
OfficeQA ProProfessional Work67.154.1−13.0
GDPval-AA v2 (normalized)Professional Work62.249.5−12.7
ALE-BenchCode24632108−12.6
VoxelBench Text RatingGames21011925−12.5
HardGeoBench <=10miMultimodal41.328.8−12.5
SWE-bench VerifiedCode95.082.6−12.4
LMArena Agent Arena Net ImprovementAgents8.04.2−12.4
HardGeoBench Avg DistanceMultimodal8001196−12.2
CursorBench 3.2Code70.558.4−12.1
SWE Atlas - Codebase QnACode33.145.2+12.1
NoCoT-Bench NCKI (kspine v3)General QA128111−12.1
LisanBench Path LengthGames136809845−11.9
ComplexConstraintsChat & Writing38.149.5+11.4
Public Benefits Bench v1 - Multi TurnProfessional Work65.253.9−11.4
Harvey's Legal Agent BenchmarkLegal13.32.1−11.2
Artificial Analysis Intelligence Index v4.3Indexes49.738.6−11.1
Document ArenaProfessional Work14971484−11.0
LLM Sycophancy Insufficient Answer RateSafety27.716.8+10.9
GDPval Extended (Mercor held-out rubric score)Professional Work65.354.5−10.8
LLM Sycophancy Contrarian RateSafety20.19.5+10.6
NoCoT-Bench NCRI 15.2General QA119100−10.3
Harvey LAB (Mercor run, 1,749 public tasks)Legal14.44.1−10.3
MMLU ProGeneral QA79.488.9+9.5
LLM Sycophancy Conditional Total InconsistencySafety26.617.1+9.5
Tax Agent BenchFinance & Accounting69.860.5−9.4
SciCode (main problems)Science21.512.3−9.2
Excel Modeling BenchmarkFinance & Accounting73.764.5−9.1
BullshitBench v2 LeaderboardSafety56.047.0−9.0
ClockBench AccuracyMultimodal37.246.1+8.9
Toloka Arena (v1)Agents60.952.3−8.6
MMMU ProMultimodal72.781.2+8.5
Furniture Assembly BenchmarkMultimodal35.844.2+8.3
APEX-AccountingFinance & Accounting56.448.2−8.2
MCP Atlas April 2026 (all 1,000 tasks)Agents83.375.3−8.0
OEIS Open Lite ($200)Math44.036.0−8.0
Artificial Analysis Agentic IndexIndexes52.844.9−7.9
TextQuests Average Cumulative HarmSafety23.721.4+7.9
SimpleQA VerifiedGeneral QA70.763.1−7.6
LLM Sycophancy Total InconsistencySafety20.613.1+7.5
OSWorld-VerifiedAgents85.077.5−7.5
Conceptual Reasoning IndexIndexes72.767.3−7.3
TextQuests Game ProgressGames56.148.9−7.2
IntPhys 2Multimodal67.159.9−7.2
WeirdML (v2)Science91.984.9−7.0
CorpFin v2 - Max Fitting ContextFinance & Accounting77.070.0−7.0
MedCodeHealth & Medicine56.149.1−7.0
BabyVision with toolsMultimodal90.583.6−6.9
Clocktower BenchGames15341486−6.9
LMCAGeneral QA61.154.3−6.9
Surface Evolver BenchAgents95.088.1−6.9
MineBench Conservative RatingGames18871772−6.8
Vision Arena OverallMultimodal13091283−6.7
DeepSearchQA (F1 score)Search94.487.8−6.6
Opus Magnum CampaignGames60.253.9−6.3
MMMU-Pro Extended (Mercor held-out)Multimodal37.643.8+6.2
ToolathlonAgents61.755.6−6.1
GeneBench-Pro Public Release SubsetScience14.08.0−6.0
TaxCalcBench TY25Finance & Accounting30.024.0−6.0
Roboflow Vision EvalsMultimodal82.276.4−5.8
FrontierFinance v3.1Finance & Accounting49.243.5−5.7
CharXiv (descriptive and reasoning combined)Multimodal89.194.6+5.5
EBR-bench v3Games39.534.3−5.2
Senior SWE-BenchCode34.729.5−5.2
Ramp SWE-Bench (78 published tasks)Code88.583.3−5.1
ArXivMath — OverallMath80.675.6−5.0
APEX-AgentsProfessional Work43.338.5−4.8
RiemannBenchMath60.055.2−4.8
LLM Creative Story WritingChat & Writing2.92.4−4.7
Machiavelli Deception (CAIS)Safety80.281.7−4.5
Finance Agent v2 (Vals Index subset)Finance & Accounting56.351.8−4.5
Finance Agent v2 (Vals Multimodal Index component)Finance & Accounting56.351.8−4.5
AutomationBench (Overall)Agents17.412.9−4.5
LiveCodeBench (Vals AI public dataset)Code89.885.3−4.5
ACCoRDGeneral QA85.080.5−4.4
SciCode (Artificial Analysis avg@3)Science60.255.9−4.3
DeepSWE v1.1Code68.664.4−4.2
Long-Horizon Terminal-Bench 1.0 (mean reward)Code48.744.5−4.2
ARC-AGI-2General QA89.285.0−4.2
LLM Sycophancy Conditional RateSafety0.64.6−4.0
Mystery Game PuzzlesGames52.056.0+4.0
CadQueryEval (Corrected Geometry)Code80.076.0−4.0
LMArena Text Arena EloChat & Writing15041481−3.9
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.960.8+3.9
ChartographyMultimodal34.831.0−3.8
GDP.pdfProfessional Work29.826.0−3.8
Terminal-Bench 2.1Code88.084.3−3.7
BrowseCompSearch88.084.4−3.6
ActiveVisionMultimodal7.110.6+3.5
ARC-AGI-1 Semi-PrivateGeneral QA98.595.0−3.5
Realm Financial ReasoningFinance & Accounting47.844.3−3.5
AA-LCRSearch70.073.3+3.3
GBENCH One-shot Coding GScoreGames65.662.2−3.3
MindCube TinyMultimodal83.179.9−3.2
LLM Sycophancy RateSafety0.53.5−3.0
ERQAMultimodal67.870.5+2.8
MASKSafety92.890.1−2.7
Agents' Last Exam (ALE)Agents25.723.0−2.6
APEX-1 (Mercor, mean score)Professional Work66.068.4+2.4
Blueprint-Bench 2Multimodal0.40.4−2.4
DTBenchGeneral QA98.496.0−2.4
Extended NYT Connections 940-Puzzle SetGames95.197.5+2.4
SUPERChemScience64.466.8+2.4
LMArena Agent Arena SteerabilityAgents11.46.9−2.3
Agent Red Teaming k=100 (CAIS)Safety39.441.5−2.2
LegalBenchLegal88.686.5−2.0
CharXiv Extended (Mercor held-out)Multimodal80.782.7+2.0
CyberGymCybersecurity83.881.8−2.0
ZeroBench Main Questions pass^5Multimodal9.011.0+2.0
CorpFin v2 - Shared Max ContextFinance & Accounting71.669.6−2.0
CorpFin v2 (Vals Index component)Finance & Accounting71.669.6−2.0
CorpFin v2 (Vals Multimodal Index component)Finance & Accounting71.669.6−2.0
TaxEval v2Finance & Accounting76.975.0−2.0
PACT Negotiation Bilateral RatingGames16031607+1.9
ChessBench AccuracyGames77.178.8+1.8
FrontierMath Tiers 1-3 (v2) - PrivateMath87.085.3−1.8
RedlineBenchLegal42.444.1+1.7
Tampered MLScience73.571.8−1.7
Artificial Analysis Coding IndexIndexes76.574.9−1.6
LLM Sycophancy Stripped RateSafety2.03.5−1.5
GPQA DiamondGeneral QA94.195.5+1.4
ZeroBench Main Mean pass@1 (five samples)Multimodal15.817.2+1.4
MineBench — Bradley-Terry RatingGames18691846−1.4
CorpFin v2 - Exact PagesFinance & Accounting66.965.6−1.3
AA-AnalystAgent (pass^5)Professional Work48.850.0+1.3
Bench to the Future 3 — Pooled ScoreForecasting0.10.1−1.2
Diligence Stack Agent BenchFinance & Accounting83.282.0−1.1
LLM Sycophancy Decisive CoverageSafety77.476.4−1.0
ProgramBench (Mercor 100-task subset)Code0.01.0+1.0
GDPval (Mercor 213-task rubric score)Professional Work83.884.6+0.8
ObviousBenchGeneral QA99.3100.0+0.7
AA-LCR Extended (Mercor held-out)Search71.170.5−0.6
τ²-bench (Telecom)Agents98.598.0−0.5
LLM Position Bias Rating BonusChat & Writing0.20.2−0.5
MedXpertQA MM (200-question set)Health & Medicine76.276.7+0.5
AutomationBench 1.0.6 (Zapier)Agents17.116.6−0.5
SAGEMath51.951.5−0.4
Ruby LLM Benchmarks — Program FixerCode77.277.6+0.4
SpatialViz-BenchMultimodal73.974.2+0.3
PCT Political Inconsistency (CAIS)Safety39.840.0−0.2
MortgageTax (Vals all 1,258 documents)Finance & Accounting68.968.8−0.2
HieroglyphBenchMultimodal22.922.8−0.1
AIME 2026Math99.9100.0+0.1
LMArena Agent Arena Bash RecoveryAgents9.69.50.0
AIME 2025Math100.0100.00.0
BabaIsBench Stage 0: The IntroGames100.0100.00.0
CRED v0.92Science96.596.50.0
FrontierMath ErdősMath0.00.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
WordleBenchGames97.097.00.0

Claude Fable 5vsGPT-5.5normalised score marginsData Nov 15, 2026Data: Mercor, Vals AI, Lech Mazur Benchmarks and 66 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…