Atlas

Compare Models

Claude Fable 5.2 vs DeepSeek V4.1 Flash 1017

Claude Fable 5.2 leads by 9.3 capability points (≈95% interval +6.5 to +12.1): a clear lead over DeepSeek V4.1 Flash 1017 · P(higher) ≈ 100%

Wins 54 of 63 shared benchmarks (9 lost)

Capability vs. relative cost
Claude Fable 5.2DeepSeek V4.1 Flash 1017Closed 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 5.2vsDeepSeek V4.1 Flash 1017capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins54 – 9 · 63 shared

Comparing matching setups only (3 benchmarks excluded) · suites counted once
← Claude Fable 5.2DeepSeek V4.1 Flash 1017 →

Claude Fable 5.2vsDeepSeek V4.1 Flash 1017wins by categoryData Nov 15, 2026Data: Artificial Analysis, Epoch AI, Arena and 15 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is DeepSeek V4.1 Flash 1017 minus Claude Fable 5.2.

ARC-AGI-3 (Semi-Private)General QA97.35.8−91.5
HardGeoBench Avg DistanceMultimodal5523201−81.7
HoneyBench v0.2 (hack rate)Safety11.976.4−64.5
Humanity's Last Exam DiamondGeneral QA80.220.8−59.4
AA-PersonalityBench Warmth indexChat & Writing82.927.7−55.8
ArXivMath 10/2026 (Native Harnesses)Math93.638.3−55.3
HardGeoBench Country AccuracyMultimodal70.015.0−55.0
FrontierMath Tier 4 (v2) - PrivateMath100.046.3−53.7
LMArena Agent Arena Net Improvement (Steering Burden)Agents30.42.0−53.5
HardGeoBench <=100miMultimodal60.07.5−52.5
SimpleQA Verified (Epoch anti-abstention)General QA94.042.5−51.5
ArXivMath 09/2026 (Native Harnesses)Math96.245.3−50.9
HardGeoBench <=10miMultimodal48.83.8−45.0
Mystery Game PuzzlesGames90.049.0−41.0
LMArena Agent Arena Praise vs ComplaintAgents73.41.2−36.1
CritPt v2 (agentic, avg@5)Science86.350.6−35.7
AA-PersonalityBench Organization indexChat & Writing54.388.6+34.6
Chess Puzzles (task version 2.0)Games79.046.0−33.0
AA-PersonalityBench Slop indexChat & Writing6.739.1+32.7
FrontierCode 2.0 - Pass RateCode54.723.0−31.7
BrokenArXiv 09/2026 (Native Harnesses)Math80.549.7−30.8
AA-PersonalityBench Imagination indexChat & Writing67.637.2−30.8
Humanity's Last ExamGeneral QA73.242.6−30.6
FrontierCode 2.0 - ScoreCode50.220.2−30.0
GameBench 2Games99.169.9−29.2
AA-PersonalityBench Expressiveness indexChat & Writing81.052.4−28.8
BuseyBench SVGMultimodal7.24.3−28.2
AA-PersonalityBench Informativeness indexChat & Writing46.771.4+25.0
FrontierMath Tiers 1-3 (v2) - PrivateMath95.171.2−23.9
BrokenArXiv 10/2026 (Native Harnesses)Math73.549.7−23.8
LMArena Code Arena WebDevCode18361664−22.5
AA-PersonalityBench Conviction indexChat & Writing75.394.3+19.2
Bug Hunt BenchCode55.237.3−18.0
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine43.926.7−17.2
ARC-AGI-2General QA93.878.1−15.7
LMArena Agent Arena Steering BurdenAgents33.04.0−14.5
AA-Briefcase v1.1 EloAgents16771416−14.3
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.741.6−14.1
FrontierCode 1.1 Main (Artificial Analysis)Code56.243.7−12.6
Epoch Capabilities Index (ECI)Indexes174159−12.0
SciCode 1.0.1 (Artificial Analysis)Code67.755.9−11.8
CritPt (avg@5)Science35.724.0−11.7
APEX-Accounting 1.1Finance & Accounting70.459.3−11.1
GDPxProfessional Work16.45.4−11.0
DAXBench v2Code88.377.7−10.6
LMArena Agent Arena Confirmed SuccessAgents22.02.3−9.8
HLAB-Vals-1.1Legal5.014.2+9.2
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal10.018.8+8.8
Chartography (with tools)Multimodal91.683.0−8.6
Excel Modeling Benchmark v1.1Finance & Accounting77.469.2−8.2
APEX-Agents 1.1Professional Work87.679.4−8.2
Tax Agent BenchFinance & Accounting67.074.9+7.9
Finance Agent v2.1Finance & Accounting69.562.9−6.5
Vibe Code Bench v1.1Code93.887.7−6.0
LMArena Agent Arena Bash ReliabilityAgents14.42.4−6.0
AA-PersonalityBench Transparency indexChat & Writing81.086.7+5.8
AutomationBench-AA 1.0.6Agents74.268.5−5.7
ARC-AGI-1 Semi-PrivateGeneral QA98.594.5−4.0
Vals Finance IndexIndexes63.660.9−2.7
DualEntry Accounting AI BenchmarkFinance & Accounting81.282.2+1.0
AA-LCR v1.1Search83.382.7−0.7
LMArena Agent Arena Tool HallucinationAgents1.00.3−0.4
OTIS Mock AIME 2024-2025Math99.799.4−0.3

Claude Fable 5.2vsDeepSeek V4.1 Flash 1017normalised score marginsData Nov 15, 2026Data: Artificial Analysis, Epoch AI, Arena and 15 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…