Atlas

Compare Models

Claude Fable 5.2 vs Claude Opus 5.5

Claude Fable 5.2 leads by 2.5 capability points (≈95% interval −0.4 to +5.4): statistically tied · P(higher) ≈ 95%

Wins 66 of 92 shared benchmarks (24 lost, 2 tied)

Capability vs. relative cost
Claude Fable 5.2Claude Opus 5.5Closed 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 5.2vsClaude Opus 5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins66 – 24 · 92 shared, 2 tied

Comparing matching setups only (4 benchmarks excluded) · suites counted once
← Claude Fable 5.2Claude Opus 5.5 →

Claude Fable 5.2vsClaude Opus 5.5wins by categoryData Nov 15, 2026Data: Artificial Analysis, Epoch AI, Arena and 29 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Claude Opus 5.5 minus Claude Fable 5.2.

AA-PersonalityBench Informativeness indexChat & Writing46.790.5+44.2
Clocktower BenchGames17591547−30.4
AA-PersonalityBench Imagination indexChat & Writing67.694.3+26.9
Humanity's Last Exam DiamondGeneral QA80.254.6−25.6
AA-PersonalityBench Organization indexChat & Writing54.379.1+25.0
SimpleQA Verified (Epoch anti-abstention)General QA94.072.2−21.8
InferenceBenchCode10.312.1+21.6
Anthropic ECI (AECI), November 2026 fitIndexes177175−20.5
LMArena Agent Arena Net Improvement (Steering Burden)Agents30.419.5−20.5
Remote Labor Index (RLI)Professional Work41.321.3−20.0
HoneyBench v0.2 (hack rate)Safety11.931.6−19.7
EBR-bench v4 (card ban, single agent)Games90.571.4−19.0
Mystery Game PuzzlesGames90.071.0−19.0
Humanity's Last Exam (Text-Only)General QA74.357.5−16.8
LMArena Agent Arena Praise vs ComplaintAgents73.440.3−16.6
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine43.927.8−16.1
Bug Hunt BenchCode55.239.7−15.6
ArXivMath 09/2026 (Native Harnesses)Math96.282.1−14.2
ArXivMath 10/2026 (Native Harnesses)Math93.679.8−13.8
SpireClimbGames19.65.8−13.8
AA-PersonalityBench Slop indexChat & Writing6.720.0+13.5
OpenWorldEngineer (Mean@8)Code54.641.6−13.0
BioSecBench-Structure V1 (58 evaluations)Science43.731.0−12.6
MineBench — Bradley-Terry RatingGames19322134+12.2
Chess Puzzles (task version 2.0)Games79.067.0−12.0
AutomationBench - release-reported metric and version unspecifiedAgents51.440.0−11.4
FrontierSWE V2 Worst@5Code62.052.0−10.1
AA-3D v1.0 EloCode12711202−10.0
AA-PersonalityBench Expressiveness indexChat & Writing81.090.5+9.6
AA-PersonalityBench Transparency indexChat & Writing81.090.5+9.6
SWE-bench Multimodal (Vals v1, 480 Tasks)Code62.370.2+7.9
FrontierSWE V2 Best@5Code79.271.5−7.7
DAXBench v2Code88.395.7+7.4
LMArena Agent Arena Steering BurdenAgents33.018.6−7.2
FrontierCode 1.1 Main (Artificial Analysis)Code64.057.1−7.0
Furniture Assembly BenchmarkMultimodal90.083.3−6.7
Humanity's Last Exam (with tools)General QA74.267.7−6.5
APEX-Agents 1.1Professional Work87.681.3−6.3
HardGeoBench <=10miMultimodal48.842.5−6.2
HieroglyphBenchMultimodal43.637.4−6.2
AA-PersonalityBench Conviction indexChat & Writing75.369.5−5.8
FrontierMath Tier 4 (v2) - PrivateMath100.095.0−5.0
APEX-Accounting 1.1Finance & Accounting70.465.5−4.9
CritPt v2 (agentic, avg@5)Science86.381.7−4.6
Epoch Capabilities Index (ECI)Indexes174169−4.5
GDPxProfessional Work16.412.0−4.4
GameBench 2Games99.194.8−4.3
BrokenArXiv 10/2026 (Native Harnesses)Math73.577.8+4.3
HLAB-Vals-1.1Legal5.09.2+4.2
HealthBenchHealth & Medicine60.164.3+4.2
ChessBenchGames69.365.3−4.0
CritPt (avg@5)Science35.731.7−4.0
FrontierMath Tiers 1-3 (v2) - PrivateMath95.191.2−3.9
BrokenArXiv 09/2026 (Native Harnesses)Math80.576.7−3.8
HardGeoBench Avg DistanceMultimodal552669−3.6
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine60.457.0−3.4
Agent Initiative BenchmarkAgents8.27.8−3.3
BuseyBench SVGMultimodal7.26.8−3.2
LMArena Code Arena WebDevCode18361814−2.8
Conceptual Reasoning IndexIndexes79.977.9−2.8
Chartography (with tools)Multimodal91.689.0−2.6
HardGeoBench Country AccuracyMultimodal70.067.5−2.5
LMCAGeneral QA70.668.2−2.4
DeepSWE v1.1Code74.372.3−2.0
AA-PersonalityBench Warmth indexChat & Writing82.984.8+1.9
Immigration BenchLegal74.072.1−1.8
Ramp Accounting BenchFinance & Accounting52.250.5−1.6
GDPval-AA v2.1Professional Work18311866+1.6
Excel Modeling Benchmark v1.1Finance & Accounting77.475.8−1.5
AA-LCR v1.1Search83.384.7+1.3
HardGeoBench <=100miMultimodal60.061.3+1.3
SWE-bench MultilingualCode96.695.5−1.1
DualEntry Accounting AI BenchmarkFinance & Accounting81.282.2+1.0
CrossSearch (pass^3)Search56.057.0+1.0
ACCoRDGeneral QA89.488.4−1.0
FrontierCode 2.0 - Pass RateCode52.751.8−0.9
FrontierCode 2.0 - ScoreCode48.247.3−0.9
AA-Briefcase v1.1 EloAgents17851801+0.9
SciCode 1.0.1 (Artificial Analysis)Code67.766.9−0.8
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal10.010.8+0.8
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.755.0−0.7
LMArena Agent Arena Confirmed SuccessAgents22.020.7−0.7
Vals RSI Index v1.1 (four tasks)Agents37.937.3−0.6
ARC-AGI-2General QA93.893.3−0.4
MacroscopeBenchCode78.578.9+0.4
LMArena Agent Arena Bash ReliabilityAgents14.415.1+0.3
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
DTBenchGeneral QA99.298.9−0.3
LMArena Agent Arena Tool HallucinationAgents1.00.7−0.2
HealthBench ProfessionalHealth & Medicine71.471.40.0
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0
ProofBench v1.1Math100.0100.00.0

Claude Fable 5.2vsClaude Opus 5.5normalised score marginsData Nov 15, 2026Data: Artificial Analysis, Epoch AI, Arena and 29 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…