Atlas

Compare Models

Claude Fable 5 vs Grok 4.6

Claude Fable 5 leads by 3.5 capability points (≈95% interval +1.3 to +5.7): a clear lead over Grok 4.6 · P(higher) ≈ 100%

Wins 111 of 142 shared benchmarks (30 lost, 1 tied)

Capability vs. relative cost
Claude Fable 5Grok 4.6Closed 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 5vsGrok 4.6capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins111 – 30 · 142 shared, 1 tied

Comparing matching setups only (65 benchmarks excluded) · suites counted once
← Claude Fable 5Grok 4.6 →

Claude Fable 5vsGrok 4.6wins by categoryData Nov 15, 2026Data: Mercor, Vals AI, Arena and 44 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Grok 4.6 minus Claude Fable 5.

NanoGPT Speedrun Frontier (human record gap closed)Agents81.710.1−96.1
KernelBench Mega Normalized (RTX PRO 6000)Code69.40.0−69.4
LLM Creative Story WritingChat & Writing2.9-2.9−64.8
Olam Social Poker EloGames15531499−58.7
FrontierMath Tier 4 (v2) - PrivateMath90.231.7−58.5
Document ArenaProfessional Work14971438−52.3
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.412.2−52.2
Antidote: Everyday EditionSafety1099966−47.5
ProofBench v1.1Math95.051.0−44.0
Hemingway-benchChat & Writing1110983−35.2
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.845.5+28.7
Vending-Bench 2Games45309047+27.4
WeirdML (v2)Science91.967.3−24.7
Humanity's Last Exam DiamondGeneral QA45.621.5−24.1
BrowseComp Extended (Mercor held-out)Search44.521.5−23.0
SAGEMath51.928.9−23.0
MirrorCode (ML, +Private, 2L)Code63.941.1−22.8
ALE-BenchCode24631822−22.7
LMArena Agent Arena Net ImprovementAgents8.01.1−22.6
ARC-AGI-2General QA89.267.1−22.1
Terminal-Bench 4.0 (Artificial Analysis)Code42.421.2−21.2
FrontierMath Tiers 1-3 (v2) - PrivateMath87.066.0−21.1
GlobeBenchGeneral QA7.328.1+20.8
SWE Atlas Codebase QnA Extended (Mercor held-out)Code0.019.3+19.3
Terminal-Bench 4.0Code44.525.3−19.2
Mystery Game PuzzlesGames52.034.0−18.0
NoCoT-Bench NCRI 15.2General QA11986.8−17.7
LLM Debate BenchmarkChat & Writing17271604−17.4
SWE Atlas - Codebase QnACode33.150.3+17.2
HealthBench Professional (Length-Adjusted)Health & Medicine60.944.4−16.5
AdvancedIFChat & Writing65.280.9+15.7
SCI-AGENT-EXAMScience67.352.4−14.9
ChartographyMultimodal34.820.5−14.3
YC-BenchGames19775731568241−14.3
LMArena Agent Arena Net Improvement (Steering Burden)Agents7.3-0.2−14.1
Vibe Code Bench v1.1Code90.476.2−14.1
Chess Puzzles (task version 2.0)Games47.033.0−14.0
Conceptual Reasoning IndexIndexes72.762.5−14.0
KernelBench CUDA Normalized (RTX PRO 6000)Code51.637.7−13.9
ChessBench CoherenceGames82.669.2−13.4
ACCoRDGeneral QA85.071.6−13.3
HLE Extended (Mercor held-out)General QA58.745.8−12.9
Extended NYT Connections 940 (Quadratic v1)Games92.780.0−12.7
LMCAGeneral QA61.148.5−12.7
Roboflow Vision Evals (October 2026, five tasks)Multimodal71.759.1−12.6
GDP.pdfProfessional Work29.817.2−12.6
Vision Arena OverallMultimodal13091263−11.9
MMMU-Pro Extended (Mercor held-out)Multimodal37.626.0−11.6
NoCoT-Bench NCKI (kspine v3)General QA128112−11.4
MedCodeHealth & Medicine56.144.7−11.4
Toloka Arena (v1)Agents60.949.7−11.2
ARC-AGI-1 Semi-PrivateGeneral QA98.587.5−11.0
Excel Modeling BenchmarkFinance & Accounting73.762.7−10.9
ReactBench (39 tasks)Code44.633.8−10.8
GBENCH One-shot Coding GScoreGames65.655.4−10.2
MMLU ProGeneral QA79.489.4+10.0
Roboflow Vision Evals (September 2026 protocol)Multimodal78.768.7−10.0
ChessBench AccuracyGames77.167.8−10.0
ChessBench EloGames1046891−9.2
EBR-bench v3Games39.530.5−9.0
HANDBOOK.mdProfessional Work36.227.3−8.9
LiveBench 2.0Agents83.674.9−8.7
Humanity's Last Exam (Text-Only)General QA51.542.9−8.6
Terminal-Bench Science 0.1 (Vals)Science12.94.3−8.6
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.948.4−8.5
Senior SWE-BenchCode34.726.3−8.4
LMArena Agent Arena Praise vs ComplaintAgents14.3-2.5−8.4
LMArena Text Arena EloChat & Writing15041456−8.3
SWE-TogetherCode68.860.6−8.2
CadQueryEval (Corrected Geometry)Code80.072.0−8.0
WordleBenchGames97.089.0−8.0
MMMU ProMultimodal72.780.5+7.8
AA-AnalystAgent (pass^5)Professional Work48.841.3−7.5
ZeroBench (Main questions, pass@5)Multimodal24.017.0−7.0
SciCode (main problems)Science21.514.6−6.9
GameBench 2Games75.268.7−6.5
Concept Synth INDUCTION Challenge64 Round 1General QA42.235.9−6.2
RedlineBenchLegal42.448.5+6.1
Concept Synth INDUCTION Challenge100 Round 1General QA31.025.0−6.0
Eval Connections One-Shot with TrapsGames85.091.0+6.0
AA-LCR Extended (Mercor held-out)Search71.165.2−5.9
TaxEval v2Finance & Accounting76.971.1−5.8
CorpFin v2Finance & Accounting71.866.2−5.7
Blueprint-Bench 2Multimodal0.40.3−5.4
Artificial Analysis Intelligence Index v4.3Indexes49.744.4−5.3
LMArena Agent Arena Steering BurdenAgents7.9-1.8−4.8
ZeroBench Main Mean pass@1 (five samples)Multimodal15.811.0−4.8
MortgageTax (Vals all 1,258 documents)Finance & Accounting68.964.2−4.7
EnterpriseBench: CoreCraftProfessional Work70.365.6−4.7
Furniture Assembly BenchmarkMultimodal35.840.0+4.2
LMArena Agent Arena SteerabilityAgents11.43.3−4.1
BabaIsBench Stage 0: The IntroGames100.096.0−4.0
LMArena Agent Arena Confirmed SuccessAgents5.1-2.6−3.9
FrontierCode 1.1 Extended - Pass RateCode70.967.0−3.8
CharXiv (descriptive and reasoning combined)Multimodal89.192.9+3.8
DAYJOB: FinanceFinance & Accounting9.012.8+3.8
VoxelBench Text RatingGames21012050−3.6
FrontierCode 1.1 Extended - ScoreCode64.961.3−3.6
MineBench — Bradley-Terry RatingGames18691811−3.5
GDPval Extended (Mercor held-out rubric score)Professional Work65.361.8−3.5
CharXiv Extended (Mercor held-out)Multimodal80.777.3−3.4
DeepSWE v1.1Code69.966.7−3.2
Arena Image-to-WebDevCode16211607−2.9
Vals RSI Index v1.1 (historical five tasks)Agents24.521.5−2.9
DAYJOB: HealthcareHealth & Medicine5.62.8−2.8
APEX-Agents 1.1Professional Work74.577.3+2.8
AA-Briefcase EloAgents15301577+2.8
RuneBenchGames6.05.8−2.7
HieroglyphBenchMultimodal22.920.3−2.6
KernelBench Hard Normalized (RTX PRO 6000)Code62.960.5−2.4
LegalBenchLegal88.686.3−2.3
GDPval (Mercor 213-task rubric score)Professional Work83.886.0+2.2
SWE-bench MultilingualCode94.892.7−2.1
ZeroBench Main Questions pass^5Multimodal9.07.0−2.0
LMArena Agent Arena Bash RecoveryAgents9.65.7−2.0
APEX-1 (Mercor, mean score)Professional Work66.067.7+1.7
ProgramBench (Mercor 100-task subset)Code0.01.7+1.7
ComplexConstraintsChat & Writing38.136.5−1.6
Tax Agent BenchFinance & Accounting69.871.4+1.6
LiveCodeBench (Vals AI public dataset)Code89.888.2−1.6
BrowseComp (Mercor 130-question subset)Search82.584.0+1.5
GDPval-AA v2Professional Work17231753+1.5
GPQA DiamondGeneral QA94.195.5+1.4
SUPERChemScience64.465.7+1.3
APEX-SWECode58.957.7−1.2
Vals RSI Index v1.1 (four tasks)Agents24.325.5+1.1
DTBenchGeneral QA98.497.3−1.1
MedXpertQA MM (200-question set)Health & Medicine76.275.2−1.0
GeneBench-Pro Public Release SubsetScience14.015.0+1.0
LMArena Code Arena WebDevCode16251618−1.0
OTIS Mock AIME 2024-2025Math100.099.2−0.8
APEX-Accounting 1.1Finance & Accounting61.060.3−0.7
SWE-bench Verified Extended (Mercor held-out)Code78.778.0−0.7
ObviousBenchGeneral QA99.3100.0+0.7
CursorBench 3.2Code70.569.9−0.6
SWE-bench VerifiedCode95.095.6+0.6
Terminal-Bench 2.1Code88.088.4+0.4
Bug Hunt BenchCode27.627.3−0.3
LMArena Agent Arena Tool HallucinationAgents0.41.0+0.3
LMArena Agent Arena Bash ReliabilityAgents10.410.0−0.2
Prophet Arena Arena Score (Fixed Context)Forecasting84.985.0+0.1
AIME 2025Math100.0100.00.0

Claude Fable 5vsGrok 4.6normalised score marginsData Nov 15, 2026Data: Mercor, Vals AI, Arena and 44 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…