Atlas

Compare Models

Claude Fable 5.2 vs Gemini 4 Argon

Claude Fable 5.2 leads by 4.3 capability points (≈95% interval +1.5 to +7.1): a clear lead over Gemini 4 Argon · P(higher) ≈ 100%

Wins 69 of 93 shared benchmarks (22 lost, 2 tied)

Capability vs. relative cost
Claude Fable 5.2Gemini 4 ArgonClosed 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 5.2vsGemini 4 Argoncapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins69 – 22 · 93 shared, 2 tied

Comparing matching setups only (3 benchmarks excluded) · suites counted once
← Claude Fable 5.2Gemini 4 Argon →

Claude Fable 5.2vsGemini 4 Argonwins by categoryData Nov 15, 2026Data: Artificial Analysis, Vals AI, Epoch AI and 28 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Argon minus Claude Fable 5.2.

AA-PersonalityBench Slop indexChat & Writing6.767.6+61.5
AA-PersonalityBench Transparency indexChat & Writing81.021.9−59.6
HoneyBench v0.2 (hack rate)Safety11.963.6−51.7
Humanity's Last Exam DiamondGeneral QA80.242.5−37.7
LMArena Agent Arena Net Improvement (Steering Burden)Agents30.410.7−37.0
AA-PersonalityBench Informativeness indexChat & Writing46.782.9+36.5
BrokenArXiv 10/2026 (Native Harnesses)Math73.539.8−33.6
BrokenArXiv 09/2026 (Native Harnesses)Math80.549.7−30.8
OpenWorldEngineer (Mean@8)Code54.624.1−30.5
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine43.913.9−30.0
BioSecBench-Structure V1 (58 evaluations)Science41.414.9−26.4
HieroglyphBenchMultimodal43.669.7+26.1
Mystery Game PuzzlesGames90.064.0−26.0
EBR-bench v4 (card ban, single agent)Games90.564.8−25.7
Agent Initiative BenchmarkAgents8.25.6−25.7
LMArena Agent Arena Praise vs ComplaintAgents73.424.1−24.7
Remote Labor Index (RLI)Professional Work41.317.1−24.2
FrontierCode 2.0 - Pass RateCode54.730.6−24.1
FrontierCode 2.0 - ScoreCode50.227.6−22.5
HLAB-Vals-1.1Legal5.027.1+22.1
FrontierMath Tier 4 (v2) - PrivateMath100.078.0−22.0
AA-PersonalityBench Imagination indexChat & Writing67.646.7−21.2
LMArena Code Arena WebDevCode18361682−20.1
Chess Puzzles (task version 2.0)Games79.059.0−20.0
AA-3D v1.0 EloCode12711147−18.1
SimpleQA Verified (Epoch anti-abstention)General QA94.076.0−18.0
Humanity's Last ExamGeneral QA73.255.6−17.6
FrontierSWE V2 Worst@5Code62.044.5−17.6
Furniture Assembly BenchmarkMultimodal90.072.5−17.5
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal10.027.4+17.4
Bug Hunt BenchCode55.238.7−16.5
MineBench — Bradley-Terry RatingGames19322195+15.9
FrontierSWE V2 Best@5Code79.264.3−14.9
FrontierCode 1.1 Main (Artificial Analysis)Code55.943.5−12.5
ACCoRDGeneral QA89.477.3−12.1
Tax Agent BenchFinance & Accounting67.079.0+12.0
AA-PersonalityBench Conviction indexChat & Writing75.363.8−11.5
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine60.449.1−11.3
CritPt v2 (agentic, avg@5)Science83.772.9−10.9
ArXivMath 10/2026 (Native Harnesses)Math93.683.0−10.6
AA-Briefcase v1.1 EloAgents16771485−10.5
Conceptual Reasoning IndexIndexes79.972.4−10.3
GameBench 2Games99.189.2−10.0
LMArena Agent Arena Bash ReliabilityAgents14.4-5.4−9.9
LMArena Agent Arena Steering BurdenAgents33.013.2−9.9
DualEntry Accounting AI BenchmarkFinance & Accounting81.291.1+9.9
FrontierMath Tiers 1-3 (v2) - PrivateMath95.186.0−9.1
GDPxProfessional Work16.47.3−9.1
Ramp Accounting BenchFinance & Accounting52.261.1+8.9
CritPt (avg@5)Science35.727.1−8.6
ArXivMath 09/2026 (Native Harnesses)Math96.287.7−8.5
CrossSearch (pass^3)Search56.064.0+8.0
BioMysteryBench (Vals Terminus 2)Science83.776.3−7.4
Vals RSI Index v1.1 (four tasks)Agents37.930.6−7.3
HealthBench ProfessionalHealth & Medicine71.464.2−7.2
LMCAGeneral QA70.663.7−6.9
MysteryMechanismScience51.845.5−6.3
HardGeoBench Country AccuracyMultimodal70.063.8−6.3
SWE-bench Multimodal (Vals v1, 480 Tasks)Code62.356.0−6.2
Epoch Capabilities Index (ECI)Indexes174167−6.2
Immigration BenchLegal74.068.1−5.9
SciCode 1.0.1 (Artificial Analysis)Code67.761.8−5.9
HardGeoBench Avg DistanceMultimodal552741−5.8
AutomationBench 1.0.6 (Zapier)Agents44.350.1+5.8
LMArena Agent Arena Confirmed SuccessAgents22.013.2−4.4
BuseyBench SVGMultimodal7.27.6+3.9
AA-PersonalityBench Warmth indexChat & Writing82.979.1−3.8
AA-PersonalityBench Expressiveness indexChat & Writing81.084.8+3.8
HardGeoBench <=10miMultimodal48.845.0−3.7
GDPval-AA v2.1Professional Work17181634−3.7
AA-LCR v1.1Search83.379.7−3.7
AutomationBench-AA 1.0.6Agents74.277.5+3.4
MacroscopeBenchCode76.072.8−3.2
Excel Modeling Benchmark v1.1Finance & Accounting77.474.8−2.5
GPQA DiamondGeneral QA93.996.4+2.5
ARC-AGI-2General QA93.891.3−2.5
Vals Finance IndexIndexes63.666.0+2.4
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.758.1+2.4
AA-PersonalityBench Organization indexChat & Writing54.356.2+1.9
Finance Agent v2.1Finance & Accounting69.567.7−1.8
ChartographyMultimodal73.471.6−1.8
DTBenchGeneral QA99.297.6−1.6
HardGeoBench <=100miMultimodal60.058.8−1.2
APEX-Accounting 1.1Finance & Accounting70.471.5+1.1
ProofBench v1.1Math100.099.0−1.0
ChessBenchGames69.370.0+0.7
ARC-AGI-3 (Semi-Private)General QA96.897.3+0.6
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
LMArena Agent Arena Tool HallucinationAgents1.00.5−0.2
APEX-Agents 1.1Professional Work87.687.4−0.2
HealthBenchHealth & Medicine60.160.0−0.1
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0
DAXBench v2Code88.388.30.0

Claude Fable 5.2vsGemini 4 Argonnormalised score marginsData Nov 15, 2026Data: Artificial Analysis, Vals AI, Epoch AI and 28 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…