Atlas

Compare Models

Gemini 4 Argon vs Gemini 4 Neon

Gemini 4 Neon leads by 3.3 capability points (≈95% interval +0.5 to +6.1): a clear lead over Gemini 4 Argon · P(higher) ≈ 99%

Wins 61 of 74 shared benchmarks (12 lost, 1 tied)

Capability vs. relative cost
Gemini 4 ArgonGemini 4 NeonClosed weightsOpen weights

Benchmark wins12 – 61 · 74 shared, 1 tied

Comparing matching setups only · suites counted once
← Gemini 4 ArgonGemini 4 Neon →
Data: Vals AI, Artificial Analysis, Surge 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 Gemini 4 Argon.

DAYJOB: AccountingProfessional › Finance, accounting & tax18.049.6+31.6
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge42.571.3+28.8
Chess Puzzles (task version 2.0)Games59.083.0+24.0
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine23.647.2+23.6
sudo L7 (October 2026 rescoring)Software › Software engineering16.138.3+22.2
Mystery Game PuzzlesGames64.085.0+21.0
Humanity's Sixth SensePerception & media › Spatial & visual reasoning55.475.9+20.5
DAYJOB: Law (November 2026 release)Professional › Legal5.926.2+20.3
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax30.550.5+20.0
FrontierMath Tier 4 (v2) - PrivateMath78.097.6+19.5
Terminal-Bench Science 0.1 (Vals)Science › Multidisciplinary44.362.9+18.6
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine13.931.7+17.8
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering25.743.3+17.6
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning72.589.2+16.7
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents10.318.4+15.8
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations36.652.3+15.7
Bug Hunt BenchSoftware › Software engineering38.754.3+15.6
Artificial Analysis Coding Agent Index v2.1Software › Software engineering45.558.6+13.1
LMArena Code Arena WebDevSoftware › Frontend & design16811764+11.1
GraphWalks BFS: 256K to 1M AggregateKnowledge & reasoning › Reading & long context84.295.2+11.0
Artificial Analysis Intelligence Index v5.0.1Composite indexes50.461.4+11.0
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents23.844.5+10.4
LiveBench 2.0 (2026-11-06 release)Composite indexes77.787.4+9.8
DAYJOB: TaxProfessional › Finance, accounting & tax3.212.8+9.6
RiemannBench (November 2026 rescoring)Math74.484.0+9.6
CursorBench 4.1Software › Software engineering56.565.4+8.9
AA-Briefcase v1.1 EloProfessional › General knowledge work14851632+8.1
SWE-Marathon v1.2Software › Software engineering54.562.5+8.0
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work25.633.4+7.8
HLAB-Vals-1.1Professional › Legal27.119.6−7.5
Vals Science IndexScience › Multidisciplinary55.462.6+7.3
SEC-Bench Pro - release-reported cohortSoftware › Cybersecurity83.490.5+7.1
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work69.075.8+6.8
ExploitGym - release-reported percentage, cohort unspecifiedSoftware › Cybersecurity42.048.3+6.3
CritPt v2 (agentic, avg@5)Science › Physical sciences71.778.0+6.3
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations50.156.3+6.2
Terminal-Bench-Science 0.1Science › Multidisciplinary48.654.8+6.2
CWE-bench v2 Pass@1Software › Cybersecurity65.070.7+5.7
FrontierMath Tiers 1-3 (v2) - PrivateMath86.091.2+5.3
GDP.xlsxProfessional › General knowledge work38.343.1+4.8
LMArena Agent Arena Steering BurdenTool & computer use › General agents12.221.2+4.5
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge46.154.8+4.3
Terminal-Bench 4.0Software › Software engineering57.661.6+4.0
Vals RSI Index v1.1 (four tasks)Software › Data & ML30.634.4+3.8
BioMysteryBench (Vals Terminus 2)Science › Life sciences76.380.0+3.7
MKA (Artificial Analysis)Professional › General knowledge work51.354.6+3.3
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents12.619.2+3.3
LVBenchPerception & media › Video understanding91.789.0−2.7
SimpleBenchKnowledge & reasoning › General knowledge91.794.4+2.7
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following70.372.9+2.6
Vals Finance IndexProfessional › Finance, accounting & tax66.068.3+2.3
APEX-1 (Mercor, mean score)Professional › General knowledge work81.178.9−2.2
ChartographyPerception & media › Documents & charts71.673.5+1.9
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations77.579.3+1.8
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax74.876.5+1.6
Vals Legal IndexProfessional › Legal76.878.4+1.6
Legal Research BenchProfessional › Legal54.856.3+1.4
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations58.159.5+1.4
APEX-Agents 1.1Professional › General knowledge work87.488.8+1.4
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering74.475.8+1.3
Finance Agent v2.1Professional › Finance, accounting & tax67.768.6+0.9
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge76.075.2−0.8
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal27.426.6−0.8
Tax Agent BenchProfessional › Finance, accounting & tax79.079.8+0.8
APEX-Accounting 1.1Professional › Finance, accounting & tax71.570.8−0.7
Gray Swan IPI Attack Success (k=15)Safety & alignment0.70.0+0.7
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering55.154.5−0.5
MysteryMechanismMath45.545.0−0.4
LMArena Text Arena EloLanguage & writing › General assistance15211518−0.4
GDP.pdf (November 2026 rescoring)Professional › General knowledge work53.653.4−0.2
GPQA DiamondScience › Multidisciplinary96.496.3−0.1
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling0.81.0+0.1
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-6.0-6.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
Data: Vals AI, Artificial Analysis, Surge 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…