Atlas

Compare Models

Gemini 3.8 Flash vs Gemini 4 Neon

Gemini 4 Neon leads by 8.3 capability points (≈95% interval +5.6 to +11.0): a clear lead over Gemini 3.8 Flash · P(higher) ≈ 100%

Wins 65 of 66 shared benchmarks (1 lost)

Capability vs. relative cost
Gemini 3.8 FlashGemini 4 NeonClosed weightsOpen weights

Benchmark wins1 – 65 · 66 shared

Comparing matching setups only (6 benchmarks excluded) · suites counted once
← Gemini 3.8 FlashGemini 4 Neon →
Data: Vals AI, Surge AI, Artificial Analysis and 10 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Neon minus Gemini 3.8 Flash.

FrontierMath Tier 4 (v2) - PrivateMath22.097.6+75.6
Vending-Bench 2Business simulations509416843+69.6
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning31.789.2+57.5
Terminal-Bench-Science 0.1Science › Multidisciplinary10.065.7+55.7
Terminal-Bench Science 0.1 (Vals)Science › Multidisciplinary8.662.9+54.3
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax4.550.5+46.0
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine2.047.2+45.2
DAYJOB: AccountingProfessional › Finance, accounting & tax6.049.6+43.6
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge29.171.3+42.2
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering19.761.6+41.9
Terminal-Bench 4.0Software › Software engineering19.761.6+41.9
Mystery Game PuzzlesGames47.085.0+38.0
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering38.275.8+37.6
Bug Hunt BenchSoftware › Software engineering17.154.3+37.1
Humanity's Sixth SensePerception & media › Spatial & visual reasoning41.675.9+34.3
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents1.518.4+33.0
ChartographyPerception & media › Documents & charts40.973.5+32.6
Artificial Analysis Intelligence Index v5.0.1Composite indexes30.861.4+30.6
LiveBench 2.0 (2026-11-06 release)Composite indexes57.387.4+30.2
sudo L7 (October 2026 rescoring)Software › Software engineering10.638.3+27.7
RiemannBench (November 2026 rescoring)Math56.884.0+27.2
Vals Science IndexScience › Multidisciplinary35.862.6+26.9
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations29.756.3+26.6
DAYJOB: Law (November 2026 release)Professional › Legal0.026.2+26.2
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations26.552.3+25.8
Chess Puzzles (task version 2.0)Games58.083.0+25.0
GDP.xlsxProfessional › General knowledge work18.343.1+24.8
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering18.743.3+24.5
LMArena Code Arena WebDevSoftware › Frontend & design15851764+23.8
AA-Briefcase v1.1 EloProfessional › General knowledge work12041632+23.5
FrontierMath Tiers 1-3 (v2) - PrivateMath68.491.2+22.8
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents4.044.5+20.2
MKA (Artificial Analysis)Professional › General knowledge work35.354.6+19.3
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations58.577.4+18.9
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations60.979.3+18.4
BioMysteryBench (Vals Terminus 2)Science › Life sciences62.280.0+17.8
Legal Research BenchProfessional › Legal38.956.3+17.3
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine15.631.7+16.1
APEX-Agents 1.1Professional › General knowledge work72.988.8+15.9
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge23.754.8+15.5
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal11.426.6+15.2
Vibe Code Bench v1.1Software › Frontend & design78.793.7+15.0
Vals Index v2.2Composite indexes56.070.9+14.9
Vals RSI Index v1.1 (four tasks)Software › Data & ML19.934.4+14.5
CritPt v2 (agentic, avg@5)Science › Physical sciences64.678.0+13.4
Tax Agent BenchProfessional › Finance, accounting & tax66.879.8+13.0
APEX-Accounting 1.1Professional › Finance, accounting & tax58.870.8+12.0
DAYJOB: TaxProfessional › Finance, accounting & tax0.812.8+12.0
SimpleBenchKnowledge & reasoning › General knowledge82.494.4+12.0
LMArena Agent Arena Steering BurdenTool & computer use › General agents-2.821.2+12.0
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations48.859.5+10.7
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work22.833.4+10.6
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following63.272.9+9.7
GDP.pdf (November 2026 rescoring)Professional › General knowledge work44.253.4+9.2
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents1.019.2+9.1
Vals Legal IndexProfessional › Legal69.878.4+8.6
MysteryMechanismMath36.545.0+8.6
Vals Finance IndexProfessional › Finance, accounting & tax60.568.3+7.9
HLAB-Vals-1.1Professional › Legal12.119.6+7.5
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents8.9-6.0−7.4
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge69.775.2+5.5
LMArena Text Arena EloLanguage & writing › General assistance14971518+3.8
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax72.976.5+3.6
Finance Agent v2.1Professional › Finance, accounting & tax65.868.6+2.8
OTIS Mock AIME 2024-2025Math98.9100.0+1.1
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling0.01.0+0.5
Data: Vals AI, Surge AI, Artificial Analysis and 10 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…