Atlas

Compare Models

Claude Fable 5 vs Claude Haiku 5.5

Claude Fable 5 leads by 2.2 capability points (≈95% interval −0.1 to +4.4): statistically tied · P(higher) ≈ 97%

Wins 73 of 99 shared benchmarks (25 lost, 1 tied)

Capability vs. relative cost
Claude Fable 5Claude 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 5vsClaude Haiku 5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins73 – 25 · 99 shared, 1 tied

Comparing matching setups only (53 benchmarks excluded) · suites counted once
← Claude Fable 5Claude Haiku 5.5 →

Claude Fable 5vsClaude Haiku 5.5wins by categoryData Nov 15, 2026Data: Mercor, Roboflow, Arena and 30 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.

KernelBench Mega Normalized (RTX PRO 6000)Code69.40.0−69.4
InferenceBenchCode8.74.4−51.4
FrontierMath Tier 4 (v2) - PrivateMath90.246.3−43.9
NoCoT-Bench NCKI (kspine v3)General QA12885.9−29.3
VoxelBench Text RatingGames21011696−28.8
TaxCalcBench TY25Finance & Accounting30.02.0−28.0
ARC-AGI-2General QA89.261.3−27.9
NoCoT-Bench NCRI 15.2General QA11968.3−27.8
Extended NYT Connections 940 (Quadratic v1)Games92.765.7−27.0
BrowseComp (Mercor 130-question subset)Search82.556.3−26.2
SWE-bench Verified Extended (Mercor held-out)Code78.753.3−25.4
Roboflow Vision Evals — Detection mAP@75 (September)Multimodal31.054.3+23.3
Mystery Game PuzzlesGames52.030.0−22.0
Roboflow Vision Evals — Detection mAP@50:95 (September)Multimodal31.453.0+21.6
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.837.7+20.9
LisanBench Difficulty-WeightedGames45622711−19.8
Eval Connections One-Shot with TrapsGames85.067.0−18.0
MineBench — Bradley-Terry RatingGames18691582−17.4
BrowseComp Extended (Mercor held-out)Search44.527.5−17.0
Roboflow Vision Evals — Identification, Strict (September)Multimodal96.980.2−16.7
ALE-BenchCode24631996−16.6
Humanity's Last Exam DiamondGeneral QA45.629.6−16.0
SWE Atlas - Codebase QnACode33.148.7+15.6
AdvancedIFChat & Writing65.280.8+15.6
Excel Modeling BenchmarkFinance & Accounting73.759.0−14.7
LMArena Text Arena EloChat & Writing15041422−14.2
Roboflow Vision Evals — Identification, Judged (September)Multimodal100.086.5−13.5
LMArena Agent Arena Net Improvement (Steering Burden)Agents7.30.5−12.9
FrontierMath Tiers 1-3 (v2) - PrivateMath87.075.1−11.9
Roboflow Vision Evals — Detection mAP@50 (September)Multimodal56.468.2+11.8
Furniture Assembly BenchmarkMultimodal35.847.5+11.7
Terminal-Bench 4.0Code44.533.3−11.2
ChessBench AccuracyGames77.166.9−10.9
KernelBench CUDA Normalized (RTX PRO 6000)Code51.641.4−10.2
ChessBench EloGames1046875−10.2
Conceptual Reasoning IndexIndexes72.765.3−10.2
LMCAGeneral QA61.151.1−10.1
GBENCH One-shot Coding GScoreGames65.655.6−9.9
SWE-bench MultilingualCode94.885.1−9.7
Terminal-Bench 4.0 (Artificial Analysis)Code42.432.8−9.6
ARC-AGI-1 Semi-PrivateGeneral QA98.589.0−9.5
Roboflow Vision Evals — Counting, Judged (September)Multimodal63.573.0+9.5
Roboflow Vision Evals — Counting, Strict (September)Multimodal63.573.0+9.5
LMArena Agent Arena Praise vs ComplaintAgents14.3-3.5−8.9
Humanity's Last Exam (Text-Only)General QA51.542.7−8.8
Roboflow Vision Evals — Reasoning, Judged (September)Multimodal66.274.8+8.6
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.948.4−8.5
MedXpertQA MM (200-question set)Health & Medicine76.267.8−8.4
Chess Puzzles (task version 2.0)Games47.055.0+8.0
MMLU ProGeneral QA79.487.4+8.0
LiveBench 2.0Agents83.675.6−8.0
Tax Agent BenchFinance & Accounting69.862.2−7.6
AA-LCR Extended (Mercor held-out)Search71.163.8−7.3
Bug Hunt BenchCode27.620.5−7.1
FrontierCode 1.1 Extended - ScoreCode64.958.1−6.8
FrontierCode 1.1 Extended - Pass RateCode70.964.1−6.7
Roboflow Vision Evals — Reasoning, Strict (September)Multimodal66.272.8+6.6
SWE-bench VerifiedCode95.588.9−6.6
MMMU ProMultimodal72.779.1+6.4
APEX-Agents 1.1Professional Work74.568.3−6.2
SAGEMath51.945.8−6.1
ProofBench v1.1Math95.089.0−6.0
MMMU-Pro Extended (Mercor held-out)Multimodal37.631.9−5.7
Vending-Bench 2Games45303602−5.6
Blueprint-Bench 2Multimodal0.40.3−5.6
AutomationBench 1.0.6 (Zapier)Agents17.112.0−5.0
AA-AnalystAgent (pass^5)Professional Work48.853.8+5.0
CharXiv (descriptive and reasoning combined)Multimodal89.194.1+5.0
MedCodeHealth & Medicine56.151.5−4.5
LMArena Code Arena WebDevCode16251594−4.1
WeirdML v3Science15.611.7−3.9
APEX-SWECode58.955.1−3.8
Harvey LAB-AA v1.0 (criterion pass rate)Legal93.689.9−3.7
ACCoRDGeneral QA85.081.5−3.4
DTBenchGeneral QA98.495.0−3.4
ProgramBench (Mercor 100-task subset)Code0.03.0+3.0
Terminal-Bench 2.1Code88.085.0−3.0
LMArena Agent Arena Steering BurdenAgents7.92.4−2.8
LisanBench Path LengthGames1368012897−2.4
SUPERChemScience64.462.0−2.4
CharXiv Extended (Mercor held-out)Multimodal80.782.9+2.2
Roboflow Vision Evals (October 2026, five tasks)Multimodal71.769.6−2.1
AA-Briefcase v1.1 EloAgents15361573+2.0
ChessBench CoherenceGames82.680.7−1.9
RedlineBenchLegal42.440.6−1.8
LMArena Agent Arena Confirmed SuccessAgents5.11.6−1.8
GDPval (Mercor 213-task rubric score)Professional Work83.885.5+1.7
LMArena Agent Arena Bash ReliabilityAgents10.47.4−1.5
APEX-1 (Mercor, mean score)Professional Work66.064.7−1.3
LMArena Agent Arena Tool HallucinationAgents0.4-2.1−1.2
APEX-Accounting 1.1Finance & Accounting61.062.2+1.2
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.463.3−1.1
OTIS Mock AIME 2024-2025Math100.098.9−1.1
BullshitBench v2 LeaderboardSafety56.057.0+1.0
GDPval Extended (Mercor held-out rubric score)Professional Work65.366.0+0.7
Prophet Arena Arena Score (Fixed Context)Forecasting84.985.4+0.5
GameBench 2Games75.275.1−0.2
Vibe Code Bench v1.1Code90.490.4+0.1
AIME 2025Math100.0100.00.0

Claude Fable 5vsClaude Haiku 5.5normalised score marginsData Nov 15, 2026Data: Mercor, Roboflow, Arena and 30 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…