Atlas

Compare Models

Gemini 4 Hydrogen vs Gemini 4 Neon

Gemini 4 Neon leads by 3.3 capability points (≈95% interval +0.4 to +6.1): a clear lead over Gemini 4 Hydrogen · P(higher) ≈ 99%

Wins 59 of 66 shared benchmarks (6 lost, 1 tied)

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

Benchmark wins6 – 59 · 66 shared, 1 tied

Comparing matching setups only (1 benchmark excluded) · suites counted once
← Gemini 4 HydrogenGemini 4 Neon →
Data: Artificial Analysis, Vals AI, Surge AI and 11 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 Hydrogen.

Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning57.589.2+31.7
Bug Hunt BenchSoftware › Software engineering31.160.0+28.9
SWE-sweepSoftware › Software engineering13.539.0+25.5
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine24.047.2+23.2
Chess Puzzles (task version 2.0)Games60.083.0+23.0
Artificial Analysis Intelligence Index v5.0.1Composite indexes41.261.4+20.2
Mystery Game PuzzlesGames65.085.0+20.0
Terminal-Bench-Science 0.1Science › Multidisciplinary51.470.5+19.1
DAYJOB: Law (November 2026 release)Professional › Legal9.726.2+16.5
DAYJOB: AccountingProfessional › Finance, accounting & tax33.649.6+16.0
Humanity's Sixth SensePerception & media › Spatial & visual reasoning63.475.9+12.5
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge58.871.3+12.5
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax38.550.5+12.0
LiveBench 2.0 (2026-11-06 release)Composite indexes77.087.4+10.4
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering53.563.6+10.1
FrontierMath Tier 4 (v2) - PrivateMath87.897.6+9.8
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents13.518.4+9.5
DAYJOB: TaxProfessional › Finance, accounting & tax3.612.8+9.2
HLAB-Vals-1.1Professional › Legal11.319.6+8.3
LMArena Code Arena WebDevSoftware › Frontend & design17031764+8.1
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations44.652.3+7.7
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents29.344.5+7.6
Terminal-Bench 4.0Software › Software engineering54.061.6+7.6
AA-Briefcase v1.1 EloProfessional › General knowledge work15691699+7.1
Artificial Analysis Legal IndexProfessional › Legal56.763.8+7.0
Artificial Analysis Finance & Accounting IndexProfessional › Finance, accounting & tax58.165.1+7.0
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering82.775.8−6.9
ChartographyPerception & media › Documents & charts66.673.5+6.9
Artificial Analysis Economics IndexComposite indexes55.061.9+6.9
Artificial Analysis Healthcare & Medical IndexHealth & medicine53.360.1+6.8
Artificial Analysis Engineering IndexScience › Engineering60.166.8+6.7
sudo L7 (October 2026 rescoring)Software › Software engineering31.738.3+6.6
Artificial Analysis Coding Agent Index v2.1Software › Software engineering52.358.6+6.3
Artificial Analysis Strategy & Ops IndexProfessional › General knowledge work57.763.8+6.0
Vibe Code Bench v1.1Software › Frontend & design87.893.7+5.9
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering39.044.4+5.5
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations51.156.3+5.2
LMArena Text Arena EloLanguage & writing › General assistance14901518+5.0
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations74.479.3+4.9
Vals Legal IndexProfessional › Legal73.678.4+4.7
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine27.231.7+4.4
SWE-Marathon v1.2Software › Software engineering58.162.5+4.4
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge70.975.2+4.3
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge46.354.8+4.3
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work71.675.8+4.2
CursorBench 4.1Software › Software engineering61.465.4+4.0
Finance Agent v2.1Professional › Finance, accounting & tax65.168.6+3.4
FrontierMath Tiers 1-3 (v2) - PrivateMath88.191.2+3.2
CritPt v2 (agentic, avg@5)Science › Physical sciences80.677.4−3.1
SimpleBenchKnowledge & reasoning › General knowledge91.794.4+2.7
CWE-bench v2 Pass@1Software › Cybersecurity68.170.7+2.6
Legal Research BenchProfessional › Legal53.856.3+2.4
Vals Index v2.2Composite indexes68.570.9+2.4
GDP.xlsxProfessional › General knowledge work40.943.1+2.2
Vals Finance IndexProfessional › Finance, accounting & tax66.168.3+2.2
LMArena Agent Arena Steering BurdenTool & computer use › General agents17.421.2+1.9
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents15.719.2+1.7
Tax Agent BenchProfessional › Finance, accounting & tax78.479.8+1.4
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following73.772.9−0.8
GDP.pdf (November 2026 rescoring)Professional › General knowledge work54.253.4−0.8
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-4.5-6.0−0.7
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax75.876.5+0.7
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work32.833.4+0.6
GPQA DiamondScience › Multidisciplinary96.896.3−0.5
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling0.71.0+0.1
OTIS Mock AIME 2024-2025Math100.0100.00.0
Data: Artificial Analysis, Vals AI, Surge AI and 11 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…