Atlas

Compare Models

Claude Fable 5 vs GPT-5.6 Sol

Claude Fable 5 leads by 0.6 capability points (≈95% interval −1.7 to +3.0): statistically tied · P(higher) ≈ 70%

Wins 138 of 216 shared benchmarks (68 lost, 10 tied)

Capability vs. relative cost
Claude Fable 5GPT-5.6 SolClosed 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-5.6 Solcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins138 – 68 · 216 shared, 10 tied

Comparing matching setups only (162 benchmarks excluded) · suites counted once
← Claude Fable 5GPT-5.6 Sol →

Claude Fable 5vsGPT-5.6 Solwins by categoryData Nov 15, 2026Data: Vals AI, OpenAI, Center for AI Safety and 77 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-5.6 Sol minus Claude Fable 5.

BridgeBench V3 Arena BullShitSafety1154868−100.0
BridgeBench V3 Arena DebuggingCode1156848−100.0
BridgeBench V3 Arena GenerationCode1101958−100.0
BridgeBench V3 Arena OverallIndexes1079934−100.0
BridgeBench V3 Arena RefactoringCode1144861−100.0
VCT-Refusal (CAIS)Safety100.011.3−88.7
HPCT Refusal (CAIS)Safety100.024.0−76.0
BridgeBench V3 Arena Speed TTFTIndexes40168873−70.8
KernelBench Mega Normalized (H100)Code78.718.6−60.1
KernelBench Hard Normalized (H100)Code99.248.6−50.5
BenchCAD - geometric overlapMultimodal38.483.4+45.0
MirrorCode (ML, +Private, 2L)Code63.920.0−43.9
YC-BenchGames1977573728001−43.6
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.426.1−38.3
LisanBench Difficulty-WeightedGames45621403−33.8
ClockBench AccuracyMultimodal37.268.3+31.1
Vending-Bench 2Games45309619+30.8
BioTIER RefusalSafety100.071.9−28.1
LisanBench Path LengthGames136804699−27.9
TextQuests Average Cumulative HarmSafety23.731.4−26.0
Antidote: Everyday EditionSafety10991027−25.7
EnterpriseBench: CoreCraftProfessional Work70.346.2−24.1
HiL-Bench (Human-in-Loop Benchmark)Agents56.332.3−24.0
Search ArenaSearch12281258+23.8
FrontierCUA pass@1Agents40.016.4−23.6
Into the Unknown Integrity Node Pass RateScience66.588.7+22.2
Furniture Assembly BenchmarkMultimodal35.856.7+20.8
Olam Social Poker EloGames15531534−20.7
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.837.4+20.6
SWE Atlas - Codebase QnACode33.153.5+20.4
BrokenArXiv 06/2026Math47.867.3+19.4
HardGeoBench Avg DistanceMultimodal8001381−17.9
SimpleBenchGeneral QA81.964.8−17.1
ActiveVisionMultimodal7.123.5+16.4
GlobeBenchGeneral QA7.323.7+16.4
HardGeoBench <=100miMultimodal56.340.0−16.2
SWE-Bench Pro (Public Dataset)Code80.364.6−15.7
Concept Synth INDUCTION Challenge64 Round 1General QA42.257.8+15.6
Humanity's Last Exam DiamondGeneral QA45.630.4−15.2
HardGeoBench <=10miMultimodal41.326.3−15.0
Document ArenaProfessional Work14971480−14.8
Hemingway-benchChat & Writing11101057−14.7
IOICode72.386.7+14.4
RiemannBenchMath60.074.4+14.4
VoxelBench Text RatingGames21012297+14.0
HardGeoBench Country AccuracyMultimodal61.347.5−13.8
τ²-bench (Telecom)Agents98.585.1−13.5
PointerbenchMultimodal78.765.3−13.4
WitnessBench v1.0Games36.423.0−13.4
Toloka Arena (v1)Agents60.947.6−13.3
HANDBOOK.mdProfessional Work36.223.5−12.7
ComplexConstraintsChat & Writing38.150.5+12.4
Humanity's Last ExamGeneral QA57.845.5−12.3
BrowseComp Extended (Mercor held-out)Search44.532.3−12.2
MedCodeHealth & Medicine56.144.0−12.1
Chess Puzzles (task version 2.0)Games47.059.0+12.0
Concept Synth INDUCTION Challenge100 Round 1General QA31.043.0+12.0
ProofBench v1.1Math95.083.0−12.0
CorpFin v2 - Max Fitting ContextFinance & Accounting77.065.2−11.9
AutomationBench 1.0.6 (Zapier)Agents17.128.8+11.7
OpenAI internal design tasks - reported scoreProfessional Work35.847.4+11.6
LLM Debate BenchmarkChat & Writing17271645−11.6
GeoBenchMultimodal44043843−11.2
SWE-TogetherCode68.857.8−11.0
APEX-SWECode58.948.0−10.9
Opus Magnum CampaignGames60.270.9+10.7
ChessBench EloGames10461218+10.2
ChartographyMultimodal34.845.0+10.2
OSWorld 2.0 - reported partial reward, patch and step limit unspecifiedAgents72.962.7−10.2
GeneBench-Pro Public Release SubsetScience14.024.0+10.0
MMMU ProMultimodal72.782.7+10.0
Vibe Code Bench v1.1Code90.480.5−9.9
ACCoRDGeneral QA85.075.9−9.0
OSWorld 2.0 - reported binary completion, patch and step limit unspecifiedAgents36.127.3−8.8
Drone-BenchAgents84.376.1−8.2
BullshitBench v2 LeaderboardSafety56.048.0−8.0
CadQueryEval (Corrected Geometry)Code80.088.0+8.0
RedlineBenchLegal42.450.3+7.9
SCI-AGENT-EXAMScience67.359.6−7.7
OpenAI internal database migration - reported scoreCode50.342.7−7.6
Machiavelli Deception (CAIS)Safety80.282.7−7.6
LiveBench 2.0Agents83.676.3−7.3
FrontierMath Tier 4 (v2) - PrivateMath90.282.9−7.3
Terminal-Bench 4.0Code44.537.3−7.2
LiveCodeBench (Vals AI public dataset)Code89.882.6−7.2
Clocktower BenchGames15341583+7.0
GDPval-AA v2Professional Work17231585−6.8
SpatialViz-BenchMultimodal73.980.7+6.8
SUPERChemScience64.471.0+6.6
ChessBench CoherenceGames82.689.1+6.5
MindCube TinyMultimodal83.176.8−6.2
ZeroBench Main Mean pass@1 (five samples)Multimodal15.822.0+6.2
Mystery Game PuzzlesGames52.058.0+6.0
AutoformBenchMath53.158.9+5.9
LMArena Agent Arena Net ImprovementAgents8.06.3−5.7
CharXiv Reasoning (with tools)Multimodal93.587.8−5.7
MaxxBenchCode63.569.1+5.6
Vision Arena OverallMultimodal13091288−5.5
Conceptual Reasoning IndexIndexes72.768.7−5.4
CorpFin v2 - Exact PagesFinance & Accounting66.961.5−5.4
Design Arena Overall Code CategoriesCode13061338+5.3
SciCode (main problems)Science21.516.2−5.3
NoCoT-Bench NCRI 15.2General QA119109−5.3
NoCoT-Bench NCKI (kspine v3)General QA128121−5.3
Diligence Stack Agent BenchFinance & Accounting83.288.3+5.2
Ramp SWE-Bench (78 published tasks)Code88.583.3−5.1
CorpFin v2 - Shared Max ContextFinance & Accounting71.666.4−5.1
CorpFin v2 (Vals Index component)Finance & Accounting71.666.4−5.1
CorpFin v2 (Vals Multimodal Index component)Finance & Accounting71.666.4−5.1
ReactBench v1Code47.552.5+5.1
LMArena Agent Arena Net Improvement (Steering Burden)Agents7.34.6−5.0
Blueprint-Bench 2Multimodal0.40.3−5.0
APEX-AccountingFinance & Accounting56.451.5−4.9
Agents' Last Exam Overall ScoreAgents48.753.6+4.9
GameBench 2Games75.280.1+4.8
HieroglyphBenchMultimodal22.927.6+4.7
Humanity's Sixth SenseMultimodal34.530.0−4.5
LMArena Agent Arena Bash RecoveryAgents9.60.7−4.4
Toolathlon-VerifiedAgents79.374.9−4.4
LLM Creative Story WritingChat & Writing2.92.5−4.3
FrontierCode 1.1 Extended - ScoreCode64.960.6−4.3
OpenAI internal data-science tasks - reported scoreScience34.730.5−4.2
TextQuests Game ProgressGames56.152.1−4.0
ZeroBench Main Questions pass^5Multimodal9.013.0+4.0
DAYJOB: HealthcareHealth & Medicine5.61.6−4.0
LMArena Agent Arena Praise vs ComplaintAgents14.36.3−4.0
Roboflow Vision EvalsMultimodal82.278.3−4.0
Into the Unknown Direction Node Pass RateScience71.074.9+3.9
ToolathlonAgents61.758.0−3.7
OpenAI internal computer-use safety - adverse outcome rateSafety18.322.0−3.7
Long-Horizon Terminal-Bench 1.0 (mean reward)Code48.745.1−3.6
DAYJOB: FinanceFinance & Accounting9.05.5−3.5
OSWorld 2.0 500-step Partial ScoreAgents66.162.6−3.5
ERQAMultimodal67.871.2+3.4
Professional Reasoning Benchmark - FinanceFinance & Accounting53.950.5−3.4
APEX-AgentsProfessional Work43.339.9−3.4
ARC-AGI-2General QA89.292.5+3.3
ChessBench — Historical Output RepairGames46.042.7−3.3
LMArena Text Arena EloChat & Writing15041485−3.3
CursorBench 3.2Code70.567.2−3.3
SciCode (Artificial Analysis avg@3)Science60.256.9−3.2
WeirdML (v2)Science91.988.8−3.2
AA-Briefcase EloAgents15301475−3.2
IntPhys 2Multimodal67.163.9−3.1
ArXivMath 06/2026Math85.488.5+3.1
BridgeBench V3 Arena Speed ThroughputIndexes10690.1−2.8
ObviousBenchGeneral QA99.396.5−2.8
DTBenchGeneral QA98.495.7−2.7
LMCAGeneral QA61.158.5−2.7
MASKSafety92.895.4+2.6
Terminal-Bench 4.0 (Artificial Analysis)Code42.439.9−2.5
CRED v0.92Science96.599.0+2.5
LMArena Agent Arena SteerabilityAgents11.46.5−2.4
FrontierFinance v3.1Finance & Accounting49.246.8−2.4
TaxEval v2Finance & Accounting76.974.8−2.2
GBENCH One-shot Coding GScoreGames65.667.7+2.1
FrontierMath Tiers 1-3 (v2) - PrivateMath87.089.1+2.1
Ruby LLM Benchmarks — Program FixerCode77.275.2−2.1
Professional Reasoning Benchmark - LegalLegal52.650.5−2.1
Eval Connections One-Shot with TrapsGames85.083.0−2.0
OSWorld-VerifiedAgents85.083.0−2.0
GSO Opt@1Code78.476.5−2.0
PCT Political Inconsistency (CAIS)Safety39.841.7−1.9
ALE-BenchCode24632410−1.9
Surface Evolver BenchAgents95.093.1−1.9
Tax Agent BenchFinance & Accounting69.868.0−1.9
Artificial Analysis Coding IndexIndexes76.578.3+1.9
MortgageTax (Vals all 1,258 documents)Finance & Accounting68.967.3−1.6
BabyVision with toolsMultimodal90.588.9−1.6
LegalBenchLegal88.687.0−1.6
Terminal-Bench 2.1Code88.089.5+1.5
LMArena Agent Arena Confirmed SuccessAgents5.12.3−1.4
Finance Agent v2 (Vals Index subset)Finance & Accounting56.355.0−1.4
Finance Agent v2 (Vals Multimodal Index component)Finance & Accounting56.355.0−1.4
Excel Modeling BenchmarkFinance & Accounting73.772.3−1.3
DeepSearchQA (F1 score)Search94.493.1−1.3
AA-AnalystAgent (pass^5)Professional Work48.847.5−1.2
Artificial Analysis Intelligence IndexIndexes62.160.9−1.2
SWE-bench VerifiedCode95.096.2+1.2
Artificial Analysis Agentic IndexIndexes52.854.0+1.2
LMArena Agent Arena Steering BurdenAgents7.95.5−1.2
BrowseCompSearch93.392.2−1.1
AA-LCRSearch70.071.0+1.0
ARC-AGI-1 Semi-PrivateGeneral QA98.597.5−1.0
Extended NYT Connections 940-Puzzle SetGames95.196.1+1.0
OEIS Open Lite ($200)Math44.043.0−1.0
ProgramBench (Mercor 100-task subset)Code0.01.0+1.0
SimpleQA VerifiedGeneral QA70.769.7−1.0
Agent Red Teaming k=100 (CAIS)Safety39.440.3−1.0
Terminal-Bench-Science 0.1Science21.422.4+1.0
EBR-bench v2Games39.538.6−1.0
GDP.pdfProfessional Work29.830.7+0.9
GPQA DiamondGeneral QA94.194.9+0.8
DeepSWE v1.1Code69.970.7+0.8
ChessBench AccuracyGames77.176.4−0.8
AutomationBench (Overall)Agents17.418.1+0.7
CyberGymCybersecurity83.884.5+0.7
SAGEMath51.952.6+0.7
GDPval-AA v2 (normalized)Professional Work62.261.6−0.6
ExploitBench - release-reported exploit progressCybersecurity78.078.5+0.5
Odysseys2Agents61.562.0+0.5
LMArena Agent Arena Bash ReliabilityAgents10.411.4+0.5
Bench to the Future 3 — Pooled ScoreForecasting0.10.1−0.5
HealthBench Professional (Length-Adjusted)Health & Medicine60.960.5−0.4
LMArena Agent Arena Tool HallucinationAgents0.41.0+0.3
Prophet Arena Arena Score (Fixed Context)Forecasting84.984.8−0.1
BabaIsBench Stage 0: The IntroGames100.0100.00.0
Eval ConnectionsGames100.0100.00.0
FrontierMath ErdősMath0.00.00.0
Into the UnknownScience0.00.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
ProofBenchMath77.077.00.0
Senior SWE-BenchCode34.734.70.0
TaxCalcBench TY25Finance & Accounting30.030.00.0
Terminal-Bench Science 0.1 (Vals)Science12.912.90.0
WordleBenchGames97.097.00.0

Claude Fable 5vsGPT-5.6 Solnormalised score marginsData Nov 15, 2026Data: Vals AI, OpenAI, Center for AI Safety and 77 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…