Atlas

Compare Models

Claude Fable 5 vs Gemini 3.7 Flash

Claude Fable 5 leads by 4.1 capability points (≈95% interval +1.9 to +6.3): a clear lead over Gemini 3.7 Flash · P(higher) ≈ 100%

Wins 111 of 156 shared benchmarks (41 lost, 4 tied)

Capability vs. relative cost
Claude Fable 5Gemini 3.7 FlashClosed 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 5vsGemini 3.7 Flashcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins111 – 41 · 156 shared, 4 tied

Comparing matching setups only (74 benchmarks excluded) · suites counted once
← Claude Fable 5Gemini 3.7 Flash →

Claude Fable 5vsGemini 3.7 Flashwins by categoryData Nov 15, 2026Data: Mercor, Arena, Center for AI Safety and 39 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 3.7 Flash minus Claude Fable 5.

VCT-Refusal (CAIS)Safety100.04.8−95.2
Olam Social Poker EloGames15531468−92.4
HPCT Refusal (CAIS)Safety100.013.0−87.0
Odysseys2Agents61.55.8−55.7
FrontierMath Tier 4 (v2) - PrivateMath90.236.6−53.6
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.415.0−49.4
SWE-bench Verified Extended (Mercor held-out)Code78.730.0−48.7
LLM Creative Story WritingChat & Writing2.9-0.7−40.2
ArXivMath 05/2026Math87.547.5−40.0
MASKSafety92.853.7−39.1
LLM Debate BenchmarkChat & Writing17271455−38.6
Terminal-Bench 4.0Code44.56.1−38.5
ProofBench v1.1Math95.058.0−37.0
BrokenArXiv 06/2026Math47.811.1−36.7
BrokenArXiv 05/2026Math44.58.5−36.0
BrokenArXiv 04/2026Math54.119.3−34.8
HieroglyphBenchMultimodal22.955.0+32.1
LMArena Agent Arena Net ImprovementAgents8.0-1.3−30.7
Terminal-Bench 4.0 (Artificial Analysis)Code42.413.6−28.8
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.844.4+27.6
HLE Extended (Mercor held-out)General QA58.731.6−27.1
HANDBOOK.mdProfessional Work36.210.8−25.4
Concept Synth INDUCTION Challenge64 Round 1General QA42.217.2−25.0
ChessBench — Historical Output RepairGames46.070.7+24.7
AA-Briefcase EloAgents15301116−24.1
ArXivMath 06/2026Math85.462.5−22.9
MirrorCode (ML, +Private, 2L)Code63.941.1−22.8
AA-LCR Extended (Mercor held-out)Search71.148.9−22.2
RiemannBenchMath60.039.2−20.8
ALE-BenchCode24631876−20.8
Senior SWE-BenchCode34.714.7−20.0
Concept Synth INDUCTION Challenge100 Round 1General QA31.011.0−20.0
Vibe Code Bench v1.1Code90.470.4−20.0
BrowseComp Extended (Mercor held-out)Search44.525.5−19.0
Chess Puzzles (task version 2.0)Games47.066.0+19.0
Document ArenaProfessional Work14971476−18.0
LMArena Agent Arena Net Improvement (Steering Burden)Agents7.3-2.2−18.0
APEX-SWECode58.942.0−16.9
LiveBench 2.0Agents83.667.2−16.4
Machiavelli Deception (CAIS)Safety80.285.5−16.1
TaxCalcBench TY25Finance & Accounting30.014.0−16.0
BioTIER RefusalSafety100.084.4−15.6
FrontierMath Tiers 1-3 (v2) - PrivateMath87.071.6−15.4
ArXivMath 04/2026Math72.557.5−15.0
Mystery Game PuzzlesGames52.037.0−15.0
SWE-bench VerifiedCode95.581.0−14.5
GlobeBenchGeneral QA7.321.8+14.5
Humanity's Last Exam DiamondGeneral QA45.631.2−14.4
EnterpriseBench: CoreCraftProfessional Work70.355.9−14.4
GDPval-AA v2Professional Work17231433−14.2
ChessBench EloGames10461265+13.0
GBENCH One-shot Coding GScoreGames65.652.6−13.0
HealthBench Professional (Length-Adjusted)Health & Medicine60.948.4−12.5
SWE-bench MultilingualCode94.882.6−12.2
Humanity's Last ExamGeneral QA57.845.6−12.2
Tax Agent BenchFinance & Accounting69.857.7−12.2
MMMU ProMultimodal72.784.9+12.2
AA-AnalystAgent (pass^5)Professional Work48.860.0+11.3
MMLU ProGeneral QA79.490.6+11.2
Conceptual Reasoning IndexIndexes72.764.6−11.1
LMCAGeneral QA61.150.4−10.8
Artificial Analysis Intelligence Index v4.3Indexes49.739.4−10.3
Arena Code WebDev — Brand & MarketingCode16231553−10.2
ERQAMultimodal67.877.7+10.0
Furniture Assembly BenchmarkMultimodal35.826.7−9.2
FrontierCode 1.1 Extended - ScoreCode64.956.3−8.7
Antidote: Everyday EditionSafety10991075−8.6
LMArena Agent Arena Praise vs ComplaintAgents14.3-2.9−8.6
FrontierCode 1.1 Extended - Pass RateCode70.962.3−8.5
Roboflow Vision Evals (October 2026, five tasks)Multimodal71.780.1+8.4
SUPERChemScience64.472.8+8.4
ACCoRDGeneral QA85.077.0−8.0
Arena Code WebDev — HTMLCode16491593−7.8
LMArena Agent Arena Steering BurdenAgents7.9-7.3−7.6
LMArena Agent Arena Bash RecoveryAgents9.6-5.4−7.5
GDPval Extended (Mercor held-out rubric score)Professional Work65.357.8−7.5
MedXpertQA MM (200-question set)Health & Medicine76.283.5+7.3
Terminal-Bench Science 0.1 (Vals)Science12.95.7−7.1
Bug Hunt BenchCode27.621.0−6.7
Arena Code WebDev — ReactCode16321584−6.5
Roboflow Vision Evals (September 2026 protocol)Multimodal78.785.2+6.5
Arena Code WebDev — GamingCode16951638−6.3
Toloka Arena (v1)Agents60.954.6−6.3
MineBench — Bradley-Terry RatingGames18691765−6.3
HardGeoBench Country AccuracyMultimodal61.367.5+6.2
Arena Code WebDev — SimulationsCode16731618−6.2
CharXiv (descriptive and reasoning combined)Multimodal89.195.3+6.2
DAYJOB: FinanceFinance & Accounting9.02.8−6.2
Arena Code WebDev — Reference-Based DesignCode16551608−6.2
Vals RSI Index v1.1 (four tasks)Agents24.318.3−6.1
GDP.pdfProfessional Work29.823.8−6.0
Arena Code WebDev — FrontendCode16371592−5.8
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.962.7+5.8
LMArena Agent Arena SteerabilityAgents11.4-0.0−5.7
ChartographyMultimodal34.840.4+5.6
DAYJOB: HealthcareHealth & Medicine5.60.0−5.6
MindCube TinyMultimodal83.188.4+5.3
PCT Political Inconsistency (CAIS)Safety39.845.1−5.3
Humanity's Sixth SenseMultimodal34.539.8+5.3
Arena Code WebDev — Content Creation ToolsCode16081574−5.3
Harvey LAB (Mercor run, 1,749 public tasks)Legal14.49.3−5.1
Eval Connections One-Shot with TrapsGames85.080.0−5.0
HardGeoBench <=10miMultimodal41.346.3+5.0
NoCoT-Bench NCKI (kspine v3)General QA128121−4.9
APEX-Agents 1.1Professional Work74.579.4+4.9
CharXiv Reasoning (with tools)Multimodal93.588.7−4.8
ARC-AGI-2General QA89.284.6−4.6
Arena Image-to-WebDevCode16211600−4.4
ComplexConstraintsChat & Writing38.142.4+4.3
LMArena Agent Arena Confirmed SuccessAgents5.1-3.5−4.3
Vision Arena OverallMultimodal13091294−4.0
CadQueryEval (Corrected Geometry)Code80.084.0+4.0
NoCoT-Bench NCRI 15.2General QA119126+3.9
APEX-Accounting 1.1Finance & Accounting61.057.2−3.8
Arena Code WebDev — FullstackCode16161629+3.6
Arena Code WebDev — Data & AnalyticsCode15531577+3.5
AIME 2025Math100.096.7−3.3
BrowseComp (Mercor 130-question subset)Search82.579.2−3.3
TextQuests Game ProgressGames56.152.9−3.2
ARC-AGI-1 Semi-PrivateGeneral QA98.595.5−3.0
HardGeoBench Avg DistanceMultimodal800895−2.9
RedlineBenchLegal42.445.3+2.9
OTIS Mock AIME 2024-2025Math100.097.2−2.8
MedCodeHealth & Medicine56.153.4−2.7
SAGEMath51.949.2−2.7
LMArena Text Arena EloChat & Writing15041489−2.5
Excel Modeling BenchmarkFinance & Accounting73.771.3−2.3
TextQuests Average Cumulative HarmSafety23.724.4−2.3
MortgageTax (Vals all 1,258 documents)Finance & Accounting68.966.7−2.3
ChessBench AccuracyGames77.179.2+2.3
Terminal-Bench 2.1Code88.085.8−2.2
TaxEval v2Finance & Accounting76.974.7−2.2
LMArena Agent Arena Bash ReliabilityAgents10.46.2−2.1
CharXiv Extended (Mercor held-out)Multimodal80.779.1−1.6
DTBenchGeneral QA98.496.8−1.6
GDPval (Mercor 213-task rubric score)Professional Work83.885.2+1.4
SpatialViz-BenchMultimodal73.975.3+1.4
Hemingway-benchChat & Writing11101105−1.4
GPQA DiamondGeneral QA94.195.5+1.4
LegalBenchLegal88.687.3−1.3
MMMU-Pro Extended (Mercor held-out)Multimodal37.636.4−1.2
IntPhys 2Multimodal67.168.2+1.1
LiveCodeBench (Vals AI public dataset)Code89.888.7−1.1
APEX-1 (Mercor, mean score)Professional Work66.067.0+1.0
ProgramBench (Mercor 100-task subset)Code0.01.0+1.0
Agent Red Teaming k=100 (CAIS)Safety39.438.5+0.9
ObviousBenchGeneral QA99.3100.0+0.7
ChessBench CoherenceGames82.683.2+0.6
AutoformBenchMath53.152.5−0.6
Arena Code WebDev — Consumer ProductCode15711574+0.5
LMArena Agent Arena Tool HallucinationAgents0.40.7+0.1
DeepSWE v1.1Code65.465.5+0.1
BabaIsBench Stage 0: The IntroGames100.0100.00.0
GeneBench-Pro Public Release SubsetScience14.014.00.0
HardGeoBench <=100miMultimodal56.356.30.0
WordleBenchGames97.097.00.0

Claude Fable 5vsGemini 3.7 Flashnormalised score marginsData Nov 15, 2026Data: Mercor, Arena, Center for AI Safety and 39 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…