Atlas

Compare Models

Claude Fable 5.2 vs Claude Haiku 5.5

Claude Fable 5.2 leads by 8.9 capability points (≈95% interval +6.3 to +11.6): a clear lead over Claude Haiku 5.5 · P(higher) ≈ 100%

Wins 79 of 82 shared benchmarks (3 lost)

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

Benchmark wins79 – 3 · 82 shared

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

Claude Fable 5.2vsClaude Haiku 5.5wins by categoryData Nov 15, 2026Data: Artificial Analysis, Vals AI, Epoch AI and 20 othersVillage LLM Atlasatlas.village.dev

Margins

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

ARC-AGI-3 (Semi-Private)General QA97.32.8−94.4
SimpleQA Verified (Epoch anti-abstention)General QA94.023.8−70.2
InferenceBenchCode10.34.4−69.7
Mystery Game PuzzlesGames90.030.0−60.0
LMArena Agent Arena Net Improvement (Steering Burden)Agents30.40.5−56.4
FrontierMath Tier 4 (v2) - PrivateMath100.046.3−53.7
Humanity's Last Exam DiamondGeneral QA80.229.6−50.6
EBR-bench v4 (card ban, single agent)Games90.543.8−46.7
Furniture Assembly BenchmarkMultimodal90.047.5−42.5
AA-3D v1.0 EloCode1271998−39.7
LMArena Agent Arena Praise vs ComplaintAgents73.4-3.5−38.5
AutomationBench-AA 1.0.6Agents74.236.0−38.2
Bug Hunt BenchCode55.220.5−34.8
ARC-AGI-2General QA93.861.3−32.5
AutomationBench 1.0.6 (Zapier)Agents44.312.0−32.3
LMArena Code Arena WebDevCode18361594−31.7
Humanity's Last Exam (Text-Only)General QA74.342.7−31.6
AA-PersonalityBench Informativeness indexChat & Writing46.775.3+28.8
ChessBenchGames69.341.3−28.0
FrontierSWE V2Code69.943.8−26.1
BuseyBench SVGMultimodal7.24.6−26.1
GameBench 2Games99.175.1−24.0
Chess Puzzles (task version 2.0)Games79.055.0−24.0
HardGeoBench Avg DistanceMultimodal5521319−23.7
FrontierCode 1.1 Main (Artificial Analysis)Code64.040.4−23.6
Agent Initiative BenchmarkAgents8.26.0−21.2
MineBench — Bradley-Terry RatingGames19321582−21.2
FrontierCode 2.0 - Pass RateCode52.731.9−20.9
Conceptual Reasoning IndexIndexes79.965.3−20.1
FrontierMath Tiers 1-3 (v2) - PrivateMath95.175.1−20.0
LMCAGeneral QA70.651.1−19.5
APEX-Agents 1.1Professional Work87.668.3−19.3
FrontierCode 2.0 - ScoreCode48.228.9−19.3
HardGeoBench Country AccuracyMultimodal70.051.3−18.8
HardGeoBench <=10miMultimodal48.830.0−18.7
CritPt v2 (agentic, avg@5)Science84.966.3−18.6
SWE-bench Multimodal (Vals v1, 480 Tasks)Code62.344.0−18.3
MysteryMechanismScience51.835.1−16.7
LMArena Agent Arena Steering BurdenAgents33.02.4−15.3
HardGeoBench <=100miMultimodal60.045.0−15.0
DeepSWE v1.1Code74.359.9−14.4
AA-PersonalityBench Organization indexChat & Writing54.341.0−13.5
AA-PersonalityBench Warmth indexChat & Writing82.969.5−13.5
BioMysteryBench (Vals Terminus 2)Science83.770.4−13.3
CritPt (avg@5)Science35.722.6−13.1
SciCode 1.0.1 (Artificial Analysis)Code67.755.0−12.7
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.743.0−12.7
Immigration BenchLegal74.061.8−12.2
Excel Modeling Benchmark v1.1Finance & Accounting77.465.5−11.9
AA-Briefcase v1.1 EloAgents17851573−11.6
SWE-bench MultilingualCode96.685.1−11.5
GDPxProfessional Work16.44.9−11.5
Epoch Capabilities Index (ECI)Indexes174160−11.3
ProofBench v1.1Math100.089.0−11.0
LMArena Agent Arena Confirmed SuccessAgents22.01.6−10.2
Ramp Accounting BenchFinance & Accounting52.242.3−9.9
ARC-AGI-1 Semi-PrivateGeneral QA98.589.0−9.5
GDPval-AA v2.1Professional Work18311629−8.9
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine43.935.0−8.9
APEX-Accounting 1.1Finance & Accounting70.462.2−8.2
ACCoRDGeneral QA89.481.5−7.9
AA-PersonalityBench Conviction indexChat & Writing75.367.6−7.7
HealthBench ProfessionalHealth & Medicine71.464.8−6.7
DAXBench v2Code88.381.9−6.4
Vals Finance IndexIndexes63.657.4−6.1
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal10.04.9−5.1
AA-LCR v1.1Search83.378.3−5.0
DualEntry Accounting AI BenchmarkFinance & Accounting81.276.2−5.0
Tax Agent BenchFinance & Accounting67.062.2−4.8
DTBenchGeneral QA99.295.0−4.2
Finance Agent v2.1Finance & Accounting69.565.4−4.1
AA-PersonalityBench Slop indexChat & Writing6.710.5+3.8
AA-PersonalityBench Imagination indexChat & Writing67.663.8−3.8
AA-PersonalityBench Expressiveness indexChat & Writing81.077.2−3.8
AA-PersonalityBench Transparency indexChat & Writing81.077.2−3.8
LMArena Agent Arena Bash ReliabilityAgents14.47.4−3.5
Vibe Code Bench v1.1Code93.890.4−3.3
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine60.458.5−1.9
LMArena Agent Arena Tool HallucinationAgents1.0-2.1−1.5
HealthBenchHealth & Medicine60.161.2+1.1
OTIS Mock AIME 2024-2025Math99.798.9−0.8
HLAB-Vals-1.1Legal5.04.2−0.8

Claude Fable 5.2vsClaude Haiku 5.5normalised score marginsData Nov 15, 2026Data: Artificial Analysis, Vals AI, Epoch AI and 20 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…