Atlas

Compare Models

Claude Fable 5 vs Gemini 3.8 Flash

Claude Fable 5 leads by 2.8 capability points (≈95% interval +0.5 to +5.1): a clear lead over Gemini 3.8 Flash · P(higher) ≈ 99%

Wins 110 of 156 shared benchmarks (41 lost, 5 tied)

Capability vs. relative cost
Claude Fable 5Gemini 3.8 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 3.8 Flashcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins110 – 41 · 156 shared, 5 tied

Comparing matching setups only (88 benchmarks excluded) · suites counted once
← Claude Fable 5Gemini 3.8 Flash →

Claude Fable 5vsGemini 3.8 Flashwins by categoryData Nov 15, 2026Data: Mercor, Center for AI Safety, Arena and 43 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 3.8 Flash minus Claude Fable 5.

VCT-Refusal (CAIS)Safety100.06.4−93.6
HPCT Refusal (CAIS)Safety100.013.0−87.0
Olam Social Poker EloGames15531479−80.4
FrontierMath Tier 4 (v2) - PrivateMath90.222.0−68.2
ProofBench v1.1Math95.048.0−47.0
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.421.7−42.8
MASKSafety92.858.6−34.2
HieroglyphBenchMultimodal22.954.6+31.7
Terminal-Bench 4.0Code44.513.1−31.4
LLM Creative Story WritingChat & Writing2.90.1−30.6
LLM Debate BenchmarkChat & Writing17271514−30.2
ArXivMath 05/2026Math87.558.3−29.2
ChessBench EloGames10461527+28.6
Concept Synth INDUCTION Challenge64 Round 1General QA42.214.1−28.1
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.844.4+27.6
BrokenArXiv 06/2026Math47.820.8−27.0
BrokenArXiv 05/2026Math44.519.0−25.5
HANDBOOK.mdProfessional Work36.210.8−25.4
SWE-bench Verified Extended (Mercor held-out)Code78.754.0−24.7
TaxCalcBench TY25Finance & Accounting30.06.0−24.0
Terminal-Bench 4.0 (Artificial Analysis)Code42.419.7−22.7
APEX-SWECode58.936.4−22.5
Concept Synth INDUCTION Challenge100 Round 1General QA31.09.0−22.0
HLE Extended (Mercor held-out)General QA58.736.8−21.9
BrokenArXiv 04/2026Math54.133.6−20.5
Machiavelli Deception (CAIS)Safety80.286.5−19.1
AA-Briefcase EloAgents15301202−19.1
FrontierMath Tiers 1-3 (v2) - PrivateMath87.068.4−18.6
Surface Evolver BenchAgents95.076.9−18.1
ArXivMath 06/2026Math85.467.4−18.1
SAGEMath51.935.1−16.8
Humanity's Last Exam DiamondGeneral QA45.629.1−16.5
LiveBench 2.0Agents83.667.3−16.3
BioTIER RefusalSafety100.083.9−16.1
CadQueryEval (Corrected Geometry)Code80.096.0+16.0
AA-LCR Extended (Mercor held-out)Search71.155.6−15.5
SWE-bench VerifiedCode95.580.6−14.9
HiL-Bench (Human-in-Loop Benchmark)Agents56.341.5−14.9
LMArena Agent Arena Net ImprovementAgents8.03.8−14.0
Document ArenaProfessional Work14971482−13.1
GDPval-AA v2Professional Work17231464−12.7
AutomationBench 1.0.6 (Zapier)Agents17.129.7+12.6
ChessBench CoherenceGames82.694.6+12.0
ERQAMultimodal67.879.6+11.8
MMMU ProMultimodal72.784.5+11.8
EnterpriseBench: CoreCraftProfessional Work70.358.5−11.8
Vibe Code Bench v1.1Code90.478.7−11.7
Humanity's Last ExamGeneral QA57.846.2−11.6
FrontierCode 1.1 Extended - ScoreCode64.953.5−11.5
MMLU ProGeneral QA79.490.7+11.3
FrontierCode 1.1 Extended - Pass RateCode70.959.6−11.2
GameBench 2Games75.264.0−11.2
Chess Puzzles (task version 2.0)Games47.058.0+11.0
Bug Hunt BenchCode27.617.1−10.5
ComplexConstraintsChat & Writing38.148.4+10.3
MaxxBenchCode63.553.4−10.1
ALE-BenchCode24632182−10.0
Conceptual Reasoning IndexIndexes72.765.7−9.6
Agent Red Teaming k=100 (CAIS)Safety39.429.8+9.5
LMArena Agent Arena Net Improvement (Steering Burden)Agents7.32.3−9.4
MindCube TinyMultimodal83.192.2+9.1
SWE-bench MultilingualCode94.885.8−9.0
RiemannBenchMath60.051.2−8.8
Artificial Analysis Intelligence Index v4.3Indexes49.741.2−8.5
MedXpertQA MM (200-question set)Health & Medicine76.284.7+8.5
ArXivMath 04/2026Math72.564.2−8.3
LMCAGeneral QA61.152.9−8.2
WeirdML v3Science15.67.4−8.2
SUPERChemScience64.472.6+8.2
Roboflow Vision Evals (October 2026, five tasks)Multimodal71.779.8+8.1
MedCodeHealth & Medicine56.148.1−7.9
SWE Atlas - Codebase QnACode33.140.6+7.5
SciCode (main problems)Science21.514.2−7.3
WeirdML (v2)Science91.984.8−7.1
Humanity's Sixth SenseMultimodal34.541.6+7.1
GBENCH One-shot Coding GScoreGames65.658.6−6.9
Vals RSI Index v1.1 (historical five tasks)Agents24.517.6−6.9
ACCoRDGeneral QA85.078.2−6.8
PLE (Physical-Engineering's Last Exam)Professional Work34.928.2−6.7
GDP.pdfProfessional Work29.823.2−6.6
PCT Political Inconsistency (CAIS)Safety39.846.3−6.5
Roboflow Vision Evals (September 2026 protocol)Multimodal78.785.1+6.4
CharXiv (descriptive and reasoning combined)Multimodal89.195.4+6.3
Physical-Engineering's Last Exam (hidden test)Professional Work34.528.2−6.3
HardGeoBench Country AccuracyMultimodal61.367.5+6.2
DAYJOB: FinanceFinance & Accounting9.02.8−6.2
ChartographyMultimodal34.840.9+6.1
Arena Image-to-WebDevCode16211594−5.8
LMArena Code Arena WebDevCode16251583−5.6
NoCoT-Bench NCKI (kspine v3)General QA128121−5.2
DeepSWE v1.1Code68.673.8+5.2
LMArena Agent Arena Bash RecoveryAgents9.6-0.7−5.1
HardGeoBench <=10miMultimodal41.346.3+5.0
Mystery Game PuzzlesGames52.047.0−5.0
GDPval Extended (Mercor held-out rubric score)Professional Work65.360.4−4.9
DAYJOB: HealthcareHealth & Medicine5.60.8−4.8
Hemingway-benchChat & Writing11101093−4.7
Extended NYT Connections 940 (Quadratic v1)Games92.797.4+4.7
APEX-AccountingFinance & Accounting56.451.7−4.7
LMArena Agent Arena Praise vs ComplaintAgents14.34.9−4.7
SWE-TogetherCode68.864.2−4.6
LMArena Agent Arena Steering BurdenAgents7.9-1.3−4.6
IntPhys 2Multimodal67.162.5−4.6
Vals RSI Index v1.1 (four tasks)Agents24.319.9−4.4
ChessBench AccuracyGames77.181.1+4.3
Terminal-Bench Science 0.1 (Vals)Science12.98.6−4.3
Antidote: Everyday EditionSafety10991087−4.3
SpatialViz-BenchMultimodal73.978.1+4.2
Vision Arena OverallMultimodal13091293−4.2
Furniture Assembly BenchmarkMultimodal35.831.7−4.2
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.960.8+3.9
MineBench — Bradley-Terry RatingGames18691806−3.8
HardGeoBench <=100miMultimodal56.360.0+3.8
NoCoT-Bench NCRI 15.2General QA119126+3.7
LMArena Agent Arena SteerabilityAgents11.44.3−3.6
MortgageTax (Vals all 1,258 documents)Finance & Accounting68.965.3−3.6
Vending-Bench 2Games45305094+3.4
AIME 2025Math100.096.7−3.3
Tax Agent BenchFinance & Accounting69.866.8−3.1
GeneBench-Pro Public Release SubsetScience14.011.0−3.0
BrowseComp Extended (Mercor held-out)Search44.541.5−3.0
Eval Connections One-Shot with TrapsGames85.088.0+3.0
APEX-1 (Mercor, mean score)Professional Work66.068.8+2.8
DTBenchGeneral QA98.495.7−2.7
AA-AnalystAgent (pass^5)Professional Work48.846.3−2.5
TaxEval v2Finance & Accounting76.974.4−2.5
TextQuests Average Cumulative HarmSafety23.724.4−2.3
TextQuests Game ProgressGames56.154.1−2.0
LMArena Agent Arena Confirmed SuccessAgents5.11.3−1.9
CharXiv Extended (Mercor held-out)Multimodal80.778.9−1.8
APEX-Agents 1.1Professional Work74.572.9−1.6
LegalBenchLegal88.687.0−1.6
Excel Modeling BenchmarkFinance & Accounting73.772.2−1.5
HardGeoBench Avg DistanceMultimodal800843−1.3
MMMU-Pro Extended (Mercor held-out)Multimodal37.636.4−1.2
LMArena Text Arena EloChat & Writing15041497−1.1
OTIS Mock AIME 2024-2025Math100.098.9−1.1
RedlineBenchLegal42.441.3−1.1
ProgramBench (Mercor 100-task subset)Code0.01.0+1.0
WordleBenchGames97.098.0+1.0
xbench-ScienceQAScience73.472.4−1.0
GPQA DiamondGeneral QA94.194.9+0.8
FrontierFinance v3.1Finance & Accounting49.250.0+0.8
ObviousBenchGeneral QA99.3100.0+0.7
BrowseComp (Mercor 130-question subset)Search82.583.1+0.6
Prophet Arena Arena Score (Fixed Context)Forecasting84.985.3+0.4
Terminal-Bench 2.1Code88.087.6−0.4
GDPval (Mercor 213-task rubric score)Professional Work83.883.5−0.3
LiveCodeBench (Vals AI public dataset)Code89.889.5−0.3
LMArena Agent Arena Bash ReliabilityAgents10.49.9−0.3
LMArena Agent Arena Tool HallucinationAgents0.40.40.0
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0
ARC-AGI-2General QA89.289.20.0
BabaIsBench Stage 0: The IntroGames100.0100.00.0
Blueprint-Bench 2Multimodal0.40.40.0
FrontierMath ErdősMath0.00.00.0

Claude Fable 5vsGemini 3.8 Flashnormalised score marginsData Nov 15, 2026Data: Mercor, Center for AI Safety, Arena and 43 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…