Atlas

Compare Models

Claude Fable 5 vs Opus 4.8

Claude Fable 5 leads by 3.8 capability points (≈95% interval +1.6 to +5.9): a clear lead over Opus 4.8 · P(higher) ≈ 100%

Wins 201 of 226 shared benchmarks (23 lost, 2 tied)

Capability vs. relative cost
Claude Fable 5Opus 4.8Closed 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 5vsOpus 4.8capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins201 – 23 · 226 shared, 2 tied

Comparing matching setups only (160 benchmarks excluded) · suites counted once
← Claude Fable 5Opus 4.8 →

Claude Fable 5vsOpus 4.8wins by categoryData Nov 15, 2026Data: Vals AI, Mercor, Anthropic and 79 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Opus 4.8 minus Claude Fable 5.

KernelBench Hard Normalized (H100)Code99.20.0−99.2
Firefox 147 Exploit DevelopmentCybersecurity88.48.8−79.6
InferenceBench (prompt v1.0)Code3.27.3+63.8
VCT-Refusal (CAIS)Safety100.041.2−58.8
NanoGPT Speedrun Frontier (human record gap closed)Agents81.739.4−56.8
BridgeBench V3 Arena OverallIndexes10791000−54.3
Clocktower BenchGames15341195−48.6
Olam Social Poker EloGames15531518−38.0
ExploitBench Anthropic capability coverage, three-trial unionCybersecurity78.040.0−38.0
BullshitBench v2 LeaderboardSafety56.094.0+38.0
Opus Magnum CampaignGames60.222.8−37.3
Concept Synth INDUCTION Challenge64 Round 1General QA42.26.3−35.9
FrontierMath Tier 4 (v2) - PrivateMath90.256.1−34.1
VoxelBench Text RatingGames21011627−33.8
ExploitBench Leaderboard (Mean capability score, out of 16)Cybersecurity10.85.6−32.7
GSO Opt@1Code78.447.1−31.4
SimpleQA VerifiedGeneral QA70.739.5−31.2
KernelBench CUDA Normalized (RTX PRO 6000)Code51.624.9−26.7
Antidote: Everyday EditionSafety10991030−24.6
Blueprint-Bench 2Multimodal0.40.1−24.1
SWE Atlas Codebase QnA Extended (Mercor held-out)Code0.024.0+24.0
LLM Creative Story WritingChat & Writing2.90.8−23.7
Humanity's Last Exam DiamondGeneral QA45.623.1−22.5
Voratiq Coding Agent LeaderboardCode22191974−22.1
ChessBench — Historical Output RepairGames46.024.0−22.0
HiL-Bench (Human-in-Loop Benchmark)Agents56.335.3−21.0
Terminal-Bench 4.0Code44.523.6−20.9
Terminal-Bench 4.0 (Artificial Analysis)Code42.421.7−20.7
LLM Position Bias Decisive Pair CoverageChat & Writing72.092.7+20.7
LisanBench Difficulty-WeightedGames45622694−20.0
Hemingway-benchChat & Writing11101038−19.9
Document ArenaProfessional Work14971475−19.7
PACT Negotiation Bilateral RatingGames16031562−19.2
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.445.6−18.9
OSS-Fuzz (Anthropic)Cybersecurity80.061.5−18.5
SWE Atlas - Codebase QnACode39.057.3+18.3
MindCube TinyMultimodal83.164.9−18.2
EnterpriseBench: CoreCraftProfessional Work70.352.3−18.0
ARC-AGI-2General QA89.272.1−17.1
SWE-bench MultimodalCode54.938.4−16.5
Mystery Game PuzzlesGames52.036.0−16.0
TextQuests Game ProgressGames56.140.1−16.0
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.832.7+15.9
ArXivMath — OverallMath80.664.8−15.8
BrokenArXiv — OverallMath48.833.0−15.8
DeepSWE v1.1Code69.954.4−15.5
SWE-bench Verified Extended (Mercor held-out)Code78.763.3−15.4
APEX-SWECode58.943.9−15.0
HPCT Refusal (CAIS)Safety100.085.0−15.0
RuneBenchGames6.05.0−15.0
KernelBench Mega Normalized (H100)Code78.763.8−14.9
BrowseComp Extended (Mercor held-out)Search44.530.0−14.5
HANDBOOK.mdProfessional Work36.221.9−14.3
EBR-bench v2Games39.525.2−14.3
PrinzBenchLegal56.042.0−14.1
Roboflow Vision EvalsMultimodal82.268.1−14.1
PCT Political Inconsistency (CAIS)Safety39.853.8−14.0
WitnessBench v1.0Games36.422.8−13.6
LLM Debate BenchmarkChat & Writing17271631−13.6
Bug Hunt BenchCode27.614.3−13.3
BabaIsBench Stage 1: The LakeGames93.380.0−13.3
LisanBench Path LengthGames136809463−13.1
ALE-BenchCode24632102−12.8
RiemannBenchMath60.047.2−12.8
GameBench 2Games75.262.5−12.8
GlobeBenchGeneral QA7.320.0+12.7
Frontier-Bench v0.1Agents33.821.1−12.7
AA-Briefcase EloAgents15301316−12.4
CadQueryEval (Corrected Geometry)Code80.068.0−12.0
IntPhys 2Multimodal67.155.1−12.0
MineBench Conservative RatingGames18871686−11.9
GBENCH One-shot Coding GScoreGames65.654.0−11.5
CRED v0.92Science96.585.0−11.5
OSWorld 2.0 500-step Partial ScoreAgents66.154.8−11.3
SWE-Bench Pro (Public Dataset)Code80.369.2−11.1
Humanity's Sixth SenseMultimodal34.523.5−11.0
EBR-bench v3Games39.528.6−11.0
Into the Unknown Direction Node Pass RateScience71.060.1−10.8
Machiavelli Deception (CAIS)Safety80.283.7−10.7
MMLU ProGeneral QA79.490.0+10.6
Public Benefits Bench v1 - NeitherProfessional Work63.453.1−10.3
Ramp SWE-Bench (78 published tasks)Code88.578.2−10.3
Chartography (with tools)Multimodal85.275.0−10.2
HealthBench Professional (Length-Adjusted)Health & Medicine66.055.8−10.2
SCI-AGENT-EXAMScience67.357.1−10.2
TaxCalcBench TY25Finance & Accounting30.020.0−10.0
FrontierCode 1.0 Extended - ScoreCode61.651.8−9.8
PointerbenchMultimodal78.768.9−9.8
Public Benefits Bench v1 - BothProfessional Work71.762.1−9.5
GDPval-AA v2Professional Work17231529−9.5
NoCoT-Bench NCRI 15.2General QA119102−9.3
HLE Extended (Mercor held-out)General QA58.749.4−9.3
BabyVision with toolsMultimodal90.581.2−9.3
YC-BenchGames19775731711389−9.3
TextQuests Average Cumulative HarmSafety23.721.0+9.3
FrontierCode 1.0 Extended - Pass RateCode65.356.1−9.2
LMArena Code Arena WebDevCode16251555−9.1
AutoformBenchMath53.144.0−9.1
WeirdML (v2)Science91.982.9−9.1
HieroglyphBenchMultimodal22.914.0−8.9
Public Benefits Bench v1 - Web SearchProfessional Work67.758.9−8.8
SpatialViz-BenchMultimodal73.965.2−8.7
ReactBench v1Code45.136.5−8.6
AdvancedIFChat & Writing65.273.7+8.5
APEX-AccountingFinance & Accounting56.448.0−8.4
ERQAMultimodal67.859.5−8.2
CursorBench 3.2Code70.562.3−8.2
Humanity's Last ExamGeneral QA57.849.8−8.0
ProofBenchMath77.069.0−8.0
AA-LCR Extended (Mercor held-out)Search71.163.2−7.9
Artificial Analysis Intelligence Index v4.3Indexes49.742.0−7.7
GDPval-AA v2 (normalized)Professional Work62.254.5−7.7
Agent Red Teaming k=100 (CAIS)Safety39.447.0−7.7
NoCoT-Bench NCKI (kspine v3)General QA128117−7.6
Vibe Code Bench v1.1Code90.482.7−7.6
ACCoRDGeneral QA85.077.4−7.6
Surface Evolver BenchAgents95.087.5−7.5
SWE-TogetherCode68.861.5−7.3
CorpFin v2 - Max Fitting ContextFinance & Accounting77.069.8−7.2
Revelo Code IndexCode42.835.6−7.2
FrontierMath Tiers 1-3 (v2) - PrivateMath87.080.0−7.0
ZeroBench (Main questions, pass@5)Multimodal24.017.0−7.0
LiveBench 2.0Agents83.676.8−6.8
Public Benefits Bench v1 - Multi TurnProfessional Work65.258.5−6.8
Furniture Assembly BenchmarkMultimodal35.842.5+6.7
Toloka Arena (v1)Agents60.954.3−6.6
SWE-bench VerifiedCode95.088.6−6.4
GDPval Extended (Mercor held-out rubric score)Professional Work65.358.9−6.4
MMMU-Pro Extended (Mercor held-out)Multimodal37.631.4−6.2
Vision Arena OverallMultimodal13091286−6.0
Conceptual Reasoning IndexIndexes72.768.3−6.0
ARC-AGI-1 Semi-PrivateGeneral QA98.592.5−6.0
SWE Atlas - Test WritingCode55.649.6−6.0
FrontierCode 1.1 Extended - Pass RateCode70.965.0−5.8
Diligence Stack Agent BenchFinance & Accounting83.277.3−5.8
GDP.pdfProfessional Work29.824.0−5.8
Artificial Analysis Agentic IndexIndexes52.847.2−5.6
CorpFin v2 - Shared Max ContextFinance & Accounting71.666.0−5.6
CorpFin v2 (Vals Index component)Finance & Accounting71.666.0−5.6
CorpFin v2 (Vals Multimodal Index component)Finance & Accounting71.666.0−5.6
GeoBenchMultimodal44044128−5.5
FrontierCode 1.1 Extended - ScoreCode64.959.5−5.5
Terminal-Bench 2.1Code88.082.7−5.3
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.951.6−5.3
BioTIER RefusalSafety100.094.7−5.3
ChessBench EloGames1046961−5.1
ChessBench AccuracyGames77.172.4−5.0
MineBench — Bradley-Terry RatingGames18691786−5.0
Harvey LAB (Mercor run, 1,749 public tasks)Legal14.49.4−5.0
Chess Puzzles (task version 2.0)Games47.052.0+5.0
OEIS Open Lite ($200)Math44.039.0−5.0
LegalBenchLegal88.683.6−5.0
Tax Agent BenchFinance & Accounting69.864.9−4.9
BrowseCompSearch93.388.5−4.8
ActiveVisionMultimodal7.12.4−4.7
Realm Financial ReasoningFinance & Accounting47.843.1−4.7
SciCode (main problems)Science21.516.9−4.6
Ruby LLM Benchmarks — Program FixerCode77.272.6−4.6
Excel Modeling BenchmarkFinance & Accounting73.769.4−4.3
Senior SWE-BenchCode34.730.5−4.2
FrontierFinance v3.1Finance & Accounting49.245.0−4.2
CharXiv (descriptive and reasoning combined)Multimodal89.193.2+4.1
SUPERChemScience64.460.3−4.1
τ²-bench (Telecom)Agents98.594.4−4.1
BabaIsBench Stage 0: The IntroGames100.096.0−4.0
CharXiv Extended (Mercor held-out)Multimodal80.776.7−4.0
GeneBench-Pro Public Release SubsetScience14.018.0+4.0
APEX-AgentsProfessional Work43.339.4−3.9
LMArena Text Arena EloChat & Writing15041482−3.9
AA-AnalystAgent (pass^5)Professional Work48.845.0−3.7
LMArena Agent Arena Net ImprovementAgents8.06.9−3.7
MedXpertQA MM (200-question set)Health & Medicine76.272.5−3.7
LMCAGeneral QA61.157.5−3.6
LLM Position Bias First-Shown Pick RateChat & Writing58.960.7−3.6
CharXiv Reasoning (with tools)Multimodal93.589.9−3.6
LLM Position Bias First-Option LiftChat & Writing8.910.7−3.6
DTBenchGeneral QA98.494.9−3.5
APEX-1 (Mercor, mean score)Professional Work66.062.7−3.3
Agents' Last Exam (ALE)Agents25.722.4−3.3
ProgramBench (Anthropic 166-task, 5-episode hidden-test pass rate)Code93.090.0−3.0
Tampered MLScience73.576.5+3.0
WordleBenchGames97.094.0−3.0
Harvey's Legal Agent BenchmarkLegal13.310.4−2.9
ChessBench CoherenceGames82.679.7−2.9
SAGEMath51.954.8+2.9
Terminal-Bench Science 0.1 (Vals)Science12.910.0−2.9
MedCodeHealth & Medicine56.153.2−2.9
MMMU-Pro (Vals 4-Option)Multimodal89.386.6−2.7
RedlineBenchLegal42.439.7−2.7
SpatialBench VerifiedScience69.266.6−2.6
CorpFin v2 - Exact PagesFinance & Accounting66.964.3−2.6
ComplexConstraintsChat & Writing38.135.6−2.5
Finance Agent v2 (Vals Index subset)Finance & Accounting56.353.9−2.4
Finance Agent v2 (Vals Multimodal Index component)Finance & Accounting56.353.9−2.4
Artificial Analysis Coding IndexIndexes76.574.3−2.2
DeepSearchQA (F1 score)Search94.492.2−2.2
GPQA DiamondGeneral QA94.191.9−2.2
Drone-BenchAgents84.382.3−2.1
AIME 2025Math100.098.3−1.7
OTIS Mock AIME 2024-2025Math100.098.3−1.7
MASKSafety92.894.4+1.6
OSWorld-VerifiedAgents85.083.4−1.6
OfficeQAProfessional Work79.077.6−1.4
TaxEval v2Finance & Accounting76.975.6−1.3
Into the Unknown Integrity Node Pass RateScience66.565.3−1.2
Vals RSI Index v1.1 (historical five tasks)Agents24.523.2−1.2
LMArena Agent Arena Bash RecoveryAgents9.67.2−1.2
MCP Atlas April 2026 (all 1,000 tasks)Agents83.382.2−1.1
scBenchScience59.358.2−1.1
LiveCodeBench (Vals Public) - MediumCode90.989.8−1.0
Extended NYT Connections 940-Puzzle SetGames95.194.1−1.0
MortgageTax (Vals all 1,258 documents)Finance & Accounting68.969.9+1.0
LMArena Agent Arena SteerabilityAgents11.49.6−0.9
Bench to the Future 3 — Pooled ScoreForecasting0.10.1−0.9
LLM Position Bias Rating BonusChat & Writing0.20.2−0.9
GDPval (Mercor 213-task rubric score)Professional Work83.884.5+0.7
Vending-Bench 2Games56805787+0.6
Long-Horizon Terminal-Bench 1.0 (mean reward)Code48.749.2+0.5
Vals RSI Index v1.1 (four tasks)Agents24.324.0−0.3
LiveCodeBench (Vals Public) - EasyCode98.899.1+0.3
AutomationBench (Overall)Agents17.417.2−0.2
AutomationBench 1.0.6 (Zapier)Agents17.116.9−0.2
LLM Position Bias Order Flip RateChat & Writing27.327.4−0.1
AIME 2026Math99.9100.0+0.1
Into the UnknownScience0.00.00.0
ObviousBenchGeneral QA99.399.30.0

Claude Fable 5vsOpus 4.8normalised score marginsData Nov 15, 2026Data: Vals AI, Mercor, Anthropic and 79 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…