Atlas

Compare Models

Claude Fable 5.1 vs Gemini 4 Neon

Gemini 4 Neon leads by 3.4 capability points (≈95% interval +0.6 to +6.2): a clear lead over Claude Fable 5.1 · P(higher) ≈ 99%

Wins 37 of 44 shared benchmarks (5 lost, 2 tied)

Capability vs. relative cost
Claude Fable 5.1Gemini 4 NeonClosed weightsOpen weights

Benchmark wins5 – 37 · 44 shared, 2 tied

Comparing matching setups only (16 benchmarks excluded) · suites counted once
← Claude Fable 5.1Gemini 4 Neon →
Data: Surge AI, Arena, Epoch AI and 8 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Neon minus Claude Fable 5.1.

Chess Puzzles (task version 2.0)Games48.083.0+35.0
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations22.456.3+34.0
ChartographyPerception & media › Documents & charts46.273.5+27.3
Mystery Game PuzzlesGames58.085.0+27.0
DAYJOB: AccountingProfessional › Finance, accounting & tax27.249.6+22.4
GDP.xlsxProfessional › General knowledge work23.143.1+20.0
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning70.089.2+19.2
Bug Hunt BenchSoftware › Software engineering41.060.0+19.0
APEX-Agents 1.1Professional › General knowledge work71.888.8+17.0
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine31.247.2+16.0
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax36.550.5+14.0
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine18.331.7+13.3
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering30.343.3+13.0
RiemannBench (November 2026 rescoring)Math71.284.0+12.8
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following60.472.9+12.5
FrontierMath Tier 4 (v2) - PrivateMath87.897.6+9.8
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents13.718.4+9.3
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal17.426.6+9.2
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents28.144.5+8.2
SimpleBenchKnowledge & reasoning › General knowledge86.694.4+7.8
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents9.3-6.0−7.6
sudo L7 (October 2026 rescoring)Software › Software engineering32.238.3+6.1
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations54.059.5+5.5
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations47.152.3+5.2
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge70.875.2+4.4
APEX-Accounting 1.1Professional › Finance, accounting & tax66.770.8+4.1
GDP.pdf (November 2026 rescoring)Professional › General knowledge work49.453.4+4.0
LMArena Agent Arena Steering BurdenTool & computer use › General agents13.521.2+3.8
DAYJOB: Law (November 2026 release)Professional › Legal22.526.2+3.7
LiveBench 2.0 (2026-11-06 release)Composite indexes84.087.4+3.5
DAYJOB: TaxProfessional › Finance, accounting & tax9.612.8+3.2
AA-Briefcase v1.1 EloProfessional › General knowledge work15771632+3.0
LMArena Code Arena WebDevSoftware › Frontend & design17411764+3.0
LMArena Text Arena EloLanguage & writing › General assistance15021518+2.9
MysteryMechanismMath47.745.0−2.7
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents14.219.2+2.5
Tax Agent BenchProfessional › Finance, accounting & tax77.979.8+1.9
Vals RSI Index v1.1 (four tasks)Software › Data & ML36.134.4−1.7
CritPt v2 (agentic, avg@5)Science › Physical sciences77.776.6−1.1
FrontierMath Tiers 1-3 (v2) - PrivateMath90.291.2+1.1
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax76.776.5−0.3
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.01.00.0
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations77.477.40.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
Data: Surge AI, Arena, Epoch AI and 8 othersatlas.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…