Atlas

Compare Models

Claude Fable 5.2 vs Kimi K3.1 Pro

Claude Fable 5.2 leads by 6.4 capability points (≈95% interval +3.7 to +9.1): a clear lead over Kimi K3.1 Pro · P(higher) ≈ 100%

Wins 45 of 54 shared benchmarks (8 lost, 1 tied)

Capability vs. relative cost
Claude Fable 5.2Kimi K3.1 ProClosed 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 5.2vsKimi K3.1 Procapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins45 – 8 · 54 shared, 1 tied

Comparing matching setups only (21 benchmarks excluded) · suites counted once
← Claude Fable 5.2Kimi K3.1 Pro →

Claude Fable 5.2vsKimi K3.1 Prowins by categoryData Nov 15, 2026Data: Vals AI, Epoch AI, Arena and 14 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Kimi K3.1 Pro minus Claude Fable 5.2.

HoneyBench v0.2 (hack rate)Safety11.971.8−59.9
LMArena Agent Arena Net Improvement (Steering Burden)Agents30.44.9−48.1
Mystery Game PuzzlesGames90.042.0−48.0
ArXivMath 10/2026 (Native Harnesses)Math93.645.7−47.9
OpenWorldEngineer (Mean@8)Code54.613.6−41.0
Humanity's Last Exam DiamondGeneral QA80.239.4−40.8
Furniture Assembly BenchmarkMultimodal90.055.8−34.2
FrontierMath Tier 4 (v2) - PrivateMath100.065.9−34.1
Chess Puzzles (task version 2.0)Games79.045.0−34.0
LMArena Agent Arena Praise vs ComplaintAgents73.48.0−32.7
FrontierCode 2.0 - Pass RateCode54.722.0−32.7
ArXivMath 09/2026 (Native Harnesses)Math96.264.2−32.1
SimpleQA Verified (Epoch anti-abstention)General QA94.062.2−31.8
FrontierCode 2.0 - ScoreCode50.219.9−30.3
MysteryMechanismScience51.826.1−25.7
Humanity's Last ExamGeneral QA73.251.1−22.1
Bug Hunt BenchCode55.234.3−21.0
LMArena Code Arena WebDevCode18361688−19.4
BrokenArXiv 09/2026 (Native Harnesses)Math80.563.5−17.0
FrontierMath Tiers 1-3 (v2) - PrivateMath95.178.2−16.8
Conceptual Reasoning IndexIndexes79.967.8−16.6
BrokenArXiv 10/2026 (Native Harnesses)Math73.558.0−15.4
GameBench 2Games99.184.0−15.1
LMArena Agent Arena Steering BurdenAgents33.06.7−13.1
LMCAGeneral QA70.657.6−13.0
ACCoRDGeneral QA89.477.5−11.9
APEX-Agents 1.1Professional Work87.675.9−11.7
Epoch Capabilities Index (ECI)Indexes174162−9.9
LMArena Agent Arena Confirmed SuccessAgents22.02.7−9.6
ProofBench v1.1Math100.091.0−9.0
Vibe Code Bench v1.1Code93.884.9−8.9
BioMysteryBench (Vals Terminus 2)Science83.775.6−8.1
APEX-Accounting 1.1Finance & Accounting70.462.4−8.0
Finance Agent v2.1Finance & Accounting69.561.5−8.0
Vals Finance IndexIndexes63.656.0−7.6
Excel Modeling Benchmark v1.1Finance & Accounting77.470.0−7.4
HLAB-Vals-1.1Legal5.011.7+6.7
DTBenchGeneral QA99.293.3−5.9
AA-Briefcase v1.1 EloAgents17851678−5.9
OfficeQA ProProfessional Work72.967.4−5.5
DualEntry Accounting AI BenchmarkFinance & Accounting81.286.1+4.9
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.759.8+4.1
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal10.013.8+3.8
MineBench — Bradley-Terry RatingGames19321992+3.6
LMArena Agent Arena Bash ReliabilityAgents14.47.2−3.6
Tax Agent BenchFinance & Accounting67.063.9−3.1
BuseyBench SVGMultimodal7.26.9−2.3
DAXBench v2Code88.386.2−2.1
Chartography (with tools)Multimodal91.690.2−1.4
DeepSWE v1.1Code74.375.5+1.2
AutomationBench-AA 1.0.6Agents74.273.3−0.9
GDPval-AA v2.1Professional Work18311846+0.7
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
LMArena Agent Arena Tool HallucinationAgents1.01.00.0

Claude Fable 5.2vsKimi K3.1 Pronormalised score marginsData Nov 15, 2026Data: Vals AI, Epoch AI, Arena and 14 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…