Atlas

Compare Models

Claude Fable 5.2 vs Gemini 4 Neon

Claude Fable 5.2 leads by 0.8 capability points (≈95% interval −2.3 to +3.9): statistically tied · P(higher) ≈ 70%

Wins 29 of 56 shared benchmarks (26 lost, 1 tied)

Capability vs. relative cost
Claude Fable 5.2Gemini 4 NeonClosed weightsOpen weights

Benchmark wins29 – 26 · 56 shared, 1 tied

Comparing matching setups only (2 benchmarks excluded) · suites counted once
← Claude Fable 5.2Gemini 4 Neon →
Data: Surge AI, Vals AI, Epoch AI and 14 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 5.2.

Vending-Bench 2Business simulations962816843+42.8
SWE-Marathon v1.2Software › Software engineering86.862.5−24.3
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge94.075.2−18.8
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents27.418.4−17.5
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal10.026.6+16.6
Humanity's Sixth SensePerception & media › Spatial & visual reasoning59.575.9+16.4
HLAB-Vals-1.1Professional › Legal5.019.6+14.6
sudo L7 (October 2026 rescoring)Software › Software engineering52.238.3−13.9
Tax Agent BenchProfessional › Finance, accounting & tax67.079.8+12.8
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine43.931.7−12.2
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents68.744.5−12.1
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations44.356.3+12.0
GDP.xlsxProfessional › General knowledge work32.943.1+10.2
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents13.1-6.0−9.5
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations68.277.4+9.2
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge80.271.3−8.9
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following64.572.9+8.4
LMArena Code Arena WebDevSoftware › Frontend & design18241764−8.0
CritPt v2 (agentic, avg@5)Science › Physical sciences84.977.4−7.4
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax43.750.5+6.8
MysteryMechanismMath51.845.0−6.8
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering49.143.3−5.9
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations74.279.3+5.2
Mystery Game PuzzlesGames90.085.0−5.0
Bug Hunt BenchSoftware › Software engineering55.260.0+4.8
Vals Finance IndexProfessional › Finance, accounting & tax63.668.3+4.7
DAYJOB: AccountingProfessional › Finance, accounting & tax45.249.6+4.4
CWE-bench v2 Pass@1Software › Cybersecurity75.070.7−4.3
Chess Puzzles (task version 2.0)Games79.083.0+4.0
FrontierMath Tiers 1-3 (v2) - PrivateMath95.191.2−3.9
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations55.759.5+3.8
BioMysteryBench (Vals Terminus 2)Science › Life sciences83.780.0−3.7
Vals RSI Index v1.1 (four tasks)Software › Data & ML37.934.4−3.5
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work72.875.8+3.0
DAYJOB: TaxProfessional › Finance, accounting & tax15.612.8−2.8
LiveBench 2.0 (2026-11-06 release)Composite indexes84.987.4+2.6
FrontierMath Tier 4 (v2) - PrivateMath100.097.6−2.4
GPQA DiamondScience › Multidisciplinary93.996.3+2.4
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine45.047.2+2.2
AA-Briefcase v1.1 EloProfessional › General knowledge work17371699−2.1
LMArena Agent Arena Steering BurdenTool & computer use › General agents24.521.2−1.6
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents22.519.2−1.6
GDP.pdf (November 2026 rescoring)Professional › General knowledge work51.853.4+1.6
APEX-Agents 1.1Professional › General knowledge work87.688.9+1.3
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax77.476.5−0.9
Finance Agent v2.1Professional › Finance, accounting & tax69.568.6−0.9
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning90.089.2−0.8
RiemannBench (November 2026 rescoring)Math83.284.0+0.8
SimpleBenchKnowledge & reasoning › General knowledge94.994.4−0.5
APEX-Accounting 1.1Professional › Finance, accounting & tax70.470.8+0.4
DAYJOB: Law (November 2026 release)Professional › Legal26.526.2−0.3
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.21.0−0.1
ChartographyPerception & media › Documents & charts73.473.5+0.1
Vibe Code Bench v1.1Software › Frontend & design93.893.7−0.1
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations52.352.30.0
Data: Surge AI, Vals AI, Epoch AI and 14 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…