Atlas

Compare Models

Claude Fable 5 vs DeepSeek V4.1 Flash

Claude Fable 5 leads by 4.1 capability points (≈95% interval +1.9 to +6.3): a clear lead over DeepSeek V4.1 Flash · P(higher) ≈ 100%

Wins 101 of 121 shared benchmarks (19 lost, 1 tied)

Capability vs. relative cost
Claude Fable 5DeepSeek V4.1 FlashClosed 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 5vsDeepSeek V4.1 Flashcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins101 – 19 · 121 shared, 1 tied

Comparing matching setups only (51 benchmarks excluded) · suites counted once
← Claude Fable 5DeepSeek V4.1 Flash →

Claude Fable 5vsDeepSeek V4.1 Flashwins by categoryData Nov 15, 2026Data: Arena, Mercor, Vals AI and 33 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 minus Claude Fable 5.

FrontierMath Tier 4 (v2) - PrivateMath90.226.8−63.4
Surface Evolver BenchAgents95.046.3−48.7
Medical Long Context Reasoning (MLCR-AA)Health & Medicine64.422.8−41.7
ProofBench v1.1Math95.054.0−41.0
LLM Creative Story WritingChat & Writing2.9-0.5−37.8
SWE Atlas Codebase QnA Extended (Mercor held-out)Code0.036.7+36.7
RiemannBenchMath60.024.0−36.0
FrontierCUA pass@1Agents40.06.7−33.3
ArXivMath 05/2026Math87.558.3−29.2
EnterpriseBench: CoreCraftProfessional Work70.341.5−28.8
ArXivMath 06/2026Math85.457.6−27.8
ArXivMath 04/2026Math72.545.0−27.5
Humanity's Last Exam DiamondGeneral QA45.619.1−26.5
TaxCalcBench TY25Finance & Accounting30.04.0−26.0
Concept Synth INDUCTION Challenge64 Round 1General QA42.217.2−25.0
APEX-Agents 1.1Professional Work74.550.5−24.0
ChartographyMultimodal34.811.6−23.2
ALE-BenchCode24631829−22.5
Terminal-Bench 2.1 Extended (Mercor held-out)Code16.838.4+21.6
KernelBench Mega Normalized (RTX PRO 6000)Code69.448.2−21.2
Humanity's Last ExamGeneral QA57.836.8−21.0
LLM Debate BenchmarkChat & Writing17271582−20.5
SWE Atlas - Codebase QnACode33.153.5+20.4
Concept Synth INDUCTION Challenge100 Round 1General QA31.011.0−20.0
FrontierMath Tiers 1-3 (v2) - PrivateMath87.067.4−19.6
Conceptual Reasoning IndexIndexes72.758.4−19.6
AdvancedIFChat & Writing65.284.0+18.8
GBENCH One-shot Coding GScoreGames65.648.0−17.6
ReactBench (39 tasks): Writing ReactCode64.347.0−17.4
ACCoRDGeneral QA85.068.1−16.9
Excel Modeling BenchmarkFinance & Accounting73.757.2−16.5
ARC-AGI-2General QA89.272.9−16.2
BrokenArXiv 05/2026Math44.528.5−16.0
Terminal-Bench 4.0 (Artificial Analysis)Code42.426.8−15.7
MedCodeHealth & Medicine56.141.2−14.9
LMCAGeneral QA61.147.0−14.2
AA-AnalystAgent (pass^5)Professional Work48.835.0−13.7
LMArena Agent Arena Net ImprovementAgents8.03.9−13.6
LMArena Agent Arena Work Net Improvement (Steering Burden)Agents8.12.5−13.5
HLE Extended (Mercor held-out)General QA58.745.3−13.4
Terminal-Bench 4.0Code44.531.2−13.3
LMArena Agent Arena Code Net Improvement (Steering Burden)Agents8.62.7−13.2
MedXpertQA MM (200-question set)Health & Medicine76.263.5−12.7
MMMU-Pro Extended (Mercor held-out)Multimodal37.625.0−12.6
SWE-bench VerifiedCode95.582.9−12.6
LMArena Agent Arena Chat Net Improvement (Steering Burden)Agents6.71.6−12.3
Senior SWE-BenchCode34.722.7−12.0
Eval Connections One-Shot with TrapsGames85.073.0−12.0
Vision Arena OverallMultimodal13091263−11.9
CharXiv Extended (Mercor held-out)Multimodal80.768.9−11.8
ComplexConstraintsChat & Writing38.149.7+11.6
YC-BenchGames19775731657236−11.2
SciCode (main problems)Science21.511.2−10.3
Artificial Analysis Intelligence Index v4.3Indexes49.739.5−10.2
GDP.pdfProfessional Work29.819.8−10.0
WeirdML v3Science15.65.9−9.6
Mystery Game PuzzlesGames52.043.0−9.0
FrontierCode 1.1 Extended - Pass RateCode70.962.0−8.8
MMLU ProGeneral QA79.488.2+8.8
Terminal-Bench Science 0.1 (Vals)Science12.94.3−8.6
DTBenchGeneral QA98.489.9−8.5
MedXpertQA MM Extended (Mercor held-out)Health & Medicine56.948.4−8.5
BrokenArXiv 06/2026Math47.839.4−8.5
FrontierCode 1.1 Extended - ScoreCode64.956.6−8.3
APEX-AccountingFinance & Accounting56.448.1−8.3
LMArena Agent Arena Chat Bash ReliabilityAgents15.0-1.0−8.0
SUPERChemScience64.456.7−7.7
Tax Agent BenchFinance & Accounting69.862.5−7.4
LMArena Agent Arena SteerabilityAgents11.4-3.3−7.4
APEX-SWECode58.951.9−7.0
Bug Hunt BenchCode27.620.6−7.0
KernelBench Hard Normalized (RTX PRO 6000)Code62.969.7+6.8
GameBench 2Games75.268.4−6.8
LMArena Agent Arena Code Steering BurdenAgents15.62.3−6.6
AA-Briefcase EloAgents15301424−6.2
LMArena Agent Arena Work Praise vs ComplaintAgents15.33.7−5.8
LMArena Agent Arena Code Praise vs ComplaintAgents14.53.0−5.7
LMArena Agent Arena Code Bash ReliabilityAgents11.2-0.2−5.7
Vibe Code Bench v1.1Code90.484.7−5.6
LMArena Agent Arena Work Bash ReliabilityAgents11.30.1−5.6
LiveBench 2.0Agents83.678.0−5.6
LMArena Agent Arena Work Steering BurdenAgents12.51.5−5.5
LegalBenchLegal88.683.3−5.3
LMArena Text Arena EloChat & Writing15041475−5.1
LMArena Agent Arena Chat Praise vs ComplaintAgents12.93.1−4.9
KernelBench CUDA Normalized (RTX PRO 6000)Code51.647.0−4.6
CyberGymCybersecurity83.888.1+4.3
DeepSWE v1.1Code69.974.2+4.3
SAGEMath51.947.9−4.0
ARC-AGI-1 Semi-PrivateGeneral QA98.594.5−4.0
CadQueryEval (Corrected Geometry)Code80.076.0−4.0
SWE-bench Verified Extended (Mercor held-out)Code78.782.7+4.0
SWE-bench MultilingualCode94.898.2+3.4
Extended NYT Connections 940 (Quadratic v1)Games92.789.6−3.1
BrowseComp (Mercor 130-question subset)Search82.579.4−3.1
GeneBench-Pro Public Release SubsetScience14.011.0−3.0
GDPval (Mercor 213-task rubric score)Professional Work83.886.7+2.9
MineBench — Bradley-Terry RatingGames18691822−2.8
Terminal-Bench 2.1Code88.090.6+2.6
Arena Image-to-WebDevCode16211610−2.2
LMArena Agent Arena Code Confirmed SuccessAgents2.26.6+2.2
LMArena Agent Arena Chat Steering BurdenAgents3.2-0.5−1.8
LMArena Agent Arena Work Confirmed SuccessAgents2.25.7+1.7
AIME 2025Math100.098.3−1.7
OTIS Mock AIME 2024-2025Math100.098.3−1.7
Furniture Assembly BenchmarkMultimodal35.834.2−1.7
CharXiv (descriptive and reasoning combined)Multimodal89.190.1+1.0
Chess Puzzles (task version 2.0)Games47.048.0+1.0
BullshitBench v2 LeaderboardSafety56.057.0+1.0
LMArena Agent Arena Bash RecoveryAgents9.67.9−0.9
xbench-ScienceQAScience73.472.6−0.8
LMArena Code Arena WebDevCode16251621−0.6
APEX-1 (Mercor, mean score)Professional Work66.065.5−0.5
LMArena Agent Arena Chat Confirmed SuccessAgents5.84.9−0.5
AA-LCR Extended (Mercor held-out)Search71.170.8−0.3
AutoformBenchMath53.152.8−0.3
LMArena Agent Arena Chat Tool HallucinationAgents0.50.3−0.1
LMArena Agent Arena Work Tool HallucinationAgents0.70.5−0.1
LMArena Agent Arena Code Tool HallucinationAgents0.70.5−0.1
GDPval-AA v2Professional Work163116320.0
ReactBench (39 tasks): Fixing ReactCode18.818.80.0

Claude Fable 5vsDeepSeek V4.1 Flashnormalised score marginsData Nov 15, 2026Data: Arena, Mercor, Vals AI and 33 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…