Atlas

Compare Models

Claude Fable 5 vs Claude Opus 5

Claude Opus 5 leads by 0.6 capability points (≈95% interval −1.8 to +3.0): statistically tied · P(higher) ≈ 69%

Wins 119 of 219 shared benchmarks (88 lost, 12 tied)

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

Benchmark wins88 – 119 · 219 shared, 12 tied

Comparing matching setups only (56 benchmarks excluded) · suites counted once
← Claude Fable 5Claude Opus 5 →

Claude Fable 5vsClaude Opus 5wins by categoryData Nov 15, 2026Data: Mercor, Vals AI, Anthropic and 66 othersVillage LLM Atlasatlas.village.dev

Margins

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

VCT-Refusal (CAIS)Safety100.029.4−70.6
BenchCAD - geometric overlapMultimodal38.489.9+51.5
Machiavelli Deception (CAIS)Safety80.296.5−49.2
BrokenArXiv 06/2026Math47.890.7+42.9
Vending-Bench 2Games453011182+40.3
SWE Atlas Codebase QnA Extended (Mercor held-out)Code0.040.0+40.0
Olam Social Poker EloGames15531518−38.0
NanoGPT Speedrun Frontier (human record gap closed)Agents81.753.6−37.7
Firefox 147 Exploit DevelopmentCybersecurity88.452.4−36.0
ClockBench AccuracyMultimodal37.265.5+28.3
Furniture Assembly BenchmarkMultimodal35.860.8+25.0
SWE Atlas - Codebase QnACode39.063.2+24.2
KernelBench Mega Normalized (H100)Code78.7100.0+21.3
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.838.0+21.2
IOICode72.391.7+19.4
ChessBench CoherenceGames82.663.8−18.8
Hemingway-benchChat & Writing11101047−17.5
FrontierMath Tier 4 (v2) - PrivateMath90.273.2−17.0
BullshitBench v2 LeaderboardSafety56.073.0+17.0
KernelBench Hard Normalized (H100)Code99.282.8−16.4
Antidote: Everyday EditionSafety10991055−15.7
AdvancedIFChat & Writing65.280.5+15.3
VoxelBench Text RatingGames21012300+14.2
ChessBench — Historical Output RepairGames46.032.7−13.3
ChessBench EloGames1046826−13.1
MMLU ProGeneral QA79.492.0+12.6
MMMU ProMultimodal72.784.0+11.3
Eval Connections One-Shot with TrapsGames85.074.0−11.0
SimpleQA VerifiedGeneral QA70.759.9−10.8
Document ArenaProfessional Work14971509+10.8
SUPERChemScience64.474.7+10.3
TaxCalcBench TY25Finance & Accounting30.020.0−10.0
Terminal-Bench Science 0.1 (Vals)Science12.922.9+10.0
LLM Creative Story WritingChat & Writing2.93.7+9.9
AutomationBench 1.0.6 (Zapier)Agents17.126.9+9.9
GameBench 2Games75.284.8+9.6
Arena Image-to-WebDevCode16211664+9.5
Terminal-Bench 4.0Code44.553.9+9.4
Concept Synth INDUCTION Challenge64 Round 1General QA42.232.8−9.4
PLE (Physical-Engineering's Last Exam)Professional Work33.842.6+8.8
AutoformBenchMath53.161.9+8.8
Search ArenaSearch12281239+8.7
Vals RSI Index v1.1 (four tasks)Agents24.333.0+8.7
GlobeBenchGeneral QA7.315.9+8.6
LMArena Code Arena WebDevCode16251691+8.6
Terminal-Bench-Science 0.1Science21.430.0+8.6
MineBench — Bradley-Terry RatingGames18692008+8.4
HieroglyphBenchMultimodal22.914.6−8.3
Blueprint-Bench 2Multimodal0.40.3−8.2
ExploitBench Anthropic capability coverage, three-trial unionCybersecurity78.070.0−8.0
RiemannBenchMath60.068.0+8.0
GeneBench-Pro Public Release SubsetScience14.022.0+8.0
Toloka Arena (v1)Agents60.952.9−8.0
AutomationBench (Overall)Agents17.425.1+7.7
Vals RSI Index v1.1 (historical five tasks)Agents24.532.1+7.6
Physical-Engineering's Last Exam (hidden test)Professional Work34.241.8+7.6
MedCodeHealth & Medicine56.163.6+7.5
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.964.1+7.2
Concept Synth INDUCTION Challenge100 Round 1General QA31.024.0−7.0
Into the Unknown Integrity Node Pass RateScience66.559.5−7.0
Mystery Game PuzzlesGames52.059.0+7.0
Tax Agent BenchFinance & Accounting69.876.7+6.8
GDPval-AA v2Professional Work17231861+6.8
AA-Briefcase EloAgents15301645+6.7
SWE Atlas - Test WritingCode55.662.2+6.6
Terminal-Bench 4.0 (Artificial Analysis)Code42.449.0+6.6
ChessBench AccuracyGames77.171.0−6.5
MineBench Conservative RatingGames18871995+6.4
Anthropic ECI (AECI), November 2026 fitIndexes169169+6.3
EBR-bench v2Games39.545.7+6.2
Vision Arena OverallMultimodal13091285−6.2
KernelBench CUDA Normalized (RTX PRO 6000)Code51.645.5−6.1
NoCoT-Bench NCKI (kspine v3)General QA128119−6.1
WitnessBench v1.0Games36.430.6−5.8
GDP.pdfProfessional Work29.824.0−5.8
GDPval-AA v2 (normalized)Professional Work62.267.9+5.7
ALE-BenchCode24632622+5.6
CharXiv (descriptive and reasoning combined)Multimodal89.194.6+5.5
Revelo Code IndexCode42.848.3+5.5
LisanBench Path LengthGames1368015428+5.4
LisanBench Difficulty-WeightedGames45625068+5.4
LMArena Agent Arena Net ImprovementAgents8.09.5+5.1
AA-AnalystAgent (pass^5)Professional Work48.853.8+5.0
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.459.4−5.0
SWE-TogetherCode68.863.8−5.0
APEX-SWECode58.963.7+4.8
OSWorld 2.0 - reported binary completion, patch and step limit unspecifiedAgents36.131.4−4.7
Agents' Last Exam (ALE)Agents25.730.3+4.6
OSWorld 2.0 - reported partial reward, patch and step limit unspecifiedAgents72.968.3−4.6
SciCode (Artificial Analysis avg@3)Science60.255.7−4.5
SWE-bench MultimodalCode54.959.4+4.5
OSWorld 2.0 500-step Partial ScoreAgents66.170.6+4.5
Tampered MLScience73.577.9+4.4
ProgramBench (Mercor 100-task subset)Code0.04.3+4.3
CharXiv Reasoning (with tools)Multimodal93.589.3−4.2
SciCode (main problems)Science21.517.3−4.2
Into the Unknown Direction Node Pass RateScience71.066.8−4.2
ExploitBench Leaderboard (Mean capability score, out of 16)Cybersecurity10.810.1−4.1
Chess Puzzles (task version 2.0)Games47.043.0−4.0
Humanity's Sixth SenseMultimodal34.530.5−4.0
ProofBench v1.1Math95.099.0+4.0
Agent Red Teaming k=100 (CAIS)Safety39.443.3−4.0
HANDBOOK.mdProfessional Work36.232.3−3.9
TextQuests Average Cumulative HarmSafety23.722.6+3.7
BabyVision with toolsMultimodal90.594.1+3.6
MedXpertQA MM (200-question set)Health & Medicine76.279.8+3.6
ActiveVisionMultimodal7.110.6+3.5
WeirdML v3Science15.619.0+3.4
Harvey LAB (Mercor run, 1,749 public tasks)Legal14.417.8+3.4
LMCAGeneral QA61.164.5+3.3
MMMU-Pro Extended (Mercor held-out)Multimodal37.634.3−3.3
SpatialBench VerifiedScience69.272.5+3.3
SWE-bench Verified Extended (Mercor held-out)Code78.782.0+3.3
DeepSWE v1.1Code69.973.2+3.2
Odysseys2Agents61.558.3−3.2
MortgageTax (Vals all 1,258 documents)Finance & Accounting68.972.1+3.1
SCI-AGENT-EXAMScience67.364.2−3.1
PCT Political Inconsistency (CAIS)Safety39.836.7+3.1
ERQAMultimodal67.870.8+3.1
BioMysteryBench - Human DifficultScience46.549.4+2.9
AA-LCR Extended (Mercor held-out)Search71.173.9+2.8
DAYJOB: HealthcareHealth & Medicine5.68.4+2.8
TextQuests Game ProgressGames56.153.3−2.8
Conceptual Reasoning IndexIndexes72.774.7+2.7
APEX-1 (Mercor, mean score)Professional Work66.068.6+2.6
Ruby LLM Benchmarks — Program FixerCode77.279.8+2.6
GBENCH One-shot Coding GScoreGames65.668.1+2.6
OSWorld-VerifiedAgents86.083.4−2.6
Drone-BenchAgents84.386.9+2.6
LMArena Agent Arena Confirmed SuccessAgents5.110.2+2.5
APEX-Agents 1.1Professional Work74.577.0+2.5
BrowseCompSearch93.390.8−2.5
MCP Atlas April 2026 (all 1,000 tasks)Agents83.385.8+2.5
LMArena Text Arena EloChat & Writing15041490−2.5
SAGEMath51.949.4−2.5
Artificial Analysis Agentic IndexIndexes52.855.3+2.4
GDPval Extended (Mercor held-out rubric score)Professional Work65.362.9−2.4
YC-BenchGames19775731911592−2.3
DAYJOB: FinanceFinance & Accounting9.011.3+2.3
HLE Extended (Mercor held-out)General QA58.761.0+2.3
NoCoT-Bench NCRI 15.2General QA119115−2.1
SWE-bench VerifiedCode95.097.0+2.0
ZeroBench (Main questions, pass@5)Multimodal24.026.0+2.0
ZeroBench Main Questions pass^5Multimodal9.011.0+2.0
LMArena Agent Arena Bash RecoveryAgents9.613.6+2.0
Vibe Code Bench v1.1Code90.488.4−1.9
IntPhys 2Multimodal67.165.1−1.9
Bug Hunt BenchCode27.625.7−1.9
APEX-AccountingFinance & Accounting56.454.5−1.9
SpatialViz-BenchMultimodal73.975.8+1.9
GDPval (Mercor 213-task rubric score)Professional Work83.885.6+1.8
TaxEval v2Finance & Accounting76.975.1−1.8
LMArena Agent Arena Steering BurdenAgents7.94.4−1.8
GPQA DiamondGeneral QA94.192.4−1.7
RedlineBenchLegal42.444.0+1.6
EnterpriseBench: CoreCraftProfessional Work70.368.7−1.6
Extended NYT Connections 940 (Quadratic v1)Games92.794.3+1.6
LegalBenchLegal88.687.0−1.6
ReactBench v1Code47.549.0+1.6
LiveBench 2.0Agents83.682.1−1.5
Artificial Analysis Coding IndexIndexes76.578.0+1.5
FrontierMath Tiers 1-3 (v2) - PrivateMath87.085.6−1.4
Humanity's Last ExamGeneral QA57.856.4−1.4
ZeroBench Main Mean pass@1 (five samples)Multimodal15.817.2+1.4
FrontierCode 1.1 Extended - Pass RateCode68.369.6+1.4
CorpFin v2Finance & Accounting71.873.2+1.4
scBenchScience59.360.6+1.3
Toolathlon-VerifiedAgents79.380.6+1.3
Bench to the Future 3 — Pooled ScoreForecasting0.10.1+1.3
ARC-AGI-2General QA89.290.4+1.3
MindCube TinyMultimodal83.184.3+1.3
Chartography (with tools)Multimodal85.284.0−1.2
ACCoRDGeneral QA85.083.8−1.1
OTIS Mock AIME 2024-2025Math100.098.9−1.1
BioMysteryBench - Human SolvableScience89.090.1+1.1
SWE-Bench Pro (Public Dataset)Code80.379.2−1.1
Terminal-Bench 2.1Code88.089.1+1.1
LMArena Agent Arena Praise vs ComplaintAgents14.312.2−1.1
Artificial Analysis Intelligence Index v4.3Indexes49.750.7+1.0
ARC-AGI-1 Semi-PrivateGeneral QA98.597.5−1.0
ProofBenchMath77.078.0+1.0
WordleBenchGames97.096.0−1.0
CharXiv Extended (Mercor held-out)Multimodal80.779.8−0.9
ComplexConstraintsChat & Writing38.137.3−0.8
Humanity's Last Exam DiamondGeneral QA45.644.8−0.8
Roboflow Vision Evals (October 2026, five tasks)Multimodal71.770.9−0.8
LiveCodeBench (Vals AI public dataset)Code89.889.0−0.7
ObviousBenchGeneral QA99.3100.0+0.7
HiL-Bench (Human-in-Loop Benchmark)Agents56.357.0+0.7
MASKSafety92.893.4+0.6
OSS-Fuzz (Anthropic)Cybersecurity80.079.4−0.6
DTBenchGeneral QA98.497.9−0.5
BrowseComp Extended (Mercor held-out)Search44.544.0−0.5
CursorBench 3.2Code70.570.0−0.5
RuneBenchGames5.75.7+0.4
Roboflow Vision Evals (September 2026 protocol)Multimodal78.778.3−0.4
LMArena Agent Arena SteerabilityAgents11.410.7−0.4
LLM Debate BenchmarkChat & Writing17271724−0.3
LMArena Agent Arena Bash ReliabilityAgents10.49.8−0.3
LMArena Agent Arena Tool HallucinationAgents0.40.9+0.3
Finance Agent v2 (Vals Index subset)Finance & Accounting56.356.6+0.2
Finance Agent v2 (Vals Multimodal Index component)Finance & Accounting56.356.6+0.2
DeepSearchQA (F1 score)Search94.494.6+0.2
WeirdML (v2)Science91.991.8−0.2
LMArena Agent Arena Net Improvement (Steering Burden)Agents7.37.2−0.1
Excel Modeling BenchmarkFinance & Accounting73.773.6−0.1
Harvey LAB-AA v1.0 (criterion pass rate)Legal93.693.5−0.1
AA-LCRSearch70.070.00.0
AIME 2025Math100.0100.00.0
BabaIsBench Stage 0: The IntroGames100.0100.00.0
BioTIER RefusalSafety100.0100.00.0
CadQueryEval (Corrected Geometry)Code80.080.00.0
FrontierMath ErdősMath0.00.00.0
HPCT Refusal (CAIS)Safety100.0100.00.0
Into the UnknownScience0.00.00.0
MaxxBenchCode63.563.50.0
ProgramBench (Anthropic 166-task, 5-episode hidden-test pass rate)Code93.093.00.0
Ramp SWE-Bench (78 published tasks)Code88.588.50.0
Senior SWE-BenchCode34.734.70.0

Claude Fable 5vsClaude Opus 5normalised score marginsData Nov 15, 2026Data: Mercor, Vals AI, Anthropic 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…