Atlas

Compare Models

Gemini 4 Neon vs GPT-6 Astra

Gemini 4 Neon leads by 1.8 capability points (≈95% interval −1.1 to +4.8): statistically tied · P(higher) ≈ 89%

Wins 49 of 62 shared benchmarks (11 lost, 2 tied)

Capability vs. relative cost
Gemini 4 NeonGPT-6 AstraClosed weightsOpen weights

Benchmark wins49 – 11 · 62 shared, 2 tied

Comparing matching setups only (40 benchmarks excluded) · suites counted once
← Gemini 4 NeonGPT-6 Astra →
Data: Vals AI, Surge AI, Artificial Analysis and 9 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6 Astra minus Gemini 4 Neon.

EnterpriseBench: CoreCraftTool & computer use › Customer service & operations77.442.6−34.8
Humanity's Sixth SensePerception & media › Spatial & visual reasoning75.953.6−22.3
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations52.331.4−20.9
GDP.xlsxProfessional › General knowledge work43.122.9−20.2
DAYJOB: AccountingProfessional › Finance, accounting & tax49.630.0−19.6
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine47.228.4−18.8
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal26.69.2−17.4
Legal Research BenchProfessional › Legal56.339.4−16.8
Tax Agent BenchProfessional › Finance, accounting & tax79.863.3−16.5
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations56.341.4−14.9
APEX-Agents 1.1Professional › General knowledge work88.975.1−13.8
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax50.537.3−13.2
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations59.546.7−12.8
Vals Finance IndexProfessional › Finance, accounting & tax68.355.6−12.7
HLAB-Vals-1.1Professional › Legal19.67.1−12.5
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations79.368.5−10.8
SimpleBenchKnowledge & reasoning › General knowledge94.483.6−10.8
DAYJOB: Law (November 2026 release)Professional › Legal26.236.6+10.4
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine31.741.7+10.0
Vals Legal IndexProfessional › Legal78.468.4−10.0
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning89.280.0−9.2
Chess Puzzles (task version 2.0)Games83.074.0−9.0
AA-Briefcase v1.1 EloProfessional › General knowledge work16991544−8.5
Finance Agent v2.1Professional › Finance, accounting & tax68.660.4−8.1
MysteryMechanismMath45.053.2+8.1
DAYJOB: TaxProfessional › Finance, accounting & tax12.84.8−8.0
LMArena Text Arena EloLanguage & writing › General assistance15181475−7.6
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents18.414.6−7.5
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge71.363.8−7.5
sudo L7 (October 2026 rescoring)Software › Software engineering38.331.1−7.2
MKA (Artificial Analysis)Professional › General knowledge work56.550.2−6.2
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-6.06.5+6.2
Terminal-Bench-Science 0.1Science › Multidisciplinary65.759.5−6.2
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents44.532.3−6.1
APEX-Accounting 1.1Professional › Finance, accounting & tax70.864.7−6.1
CritPt v2 (agentic, avg@5)Science › Physical sciences78.084.0+6.0
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge54.844.6−5.1
SEC-Bench Pro - release-reported cohortSoftware › Cybersecurity90.585.4−5.1
Artificial Analysis Intelligence Index v5.0.1Composite indexes61.456.9−4.5
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering63.659.6−4.0
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax76.572.7−3.8
LMArena Agent Arena Steering BurdenTool & computer use › General agents21.213.9−3.6
LiveBench 2.0 (2026-11-06 release)Composite indexes87.484.3−3.2
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering44.441.4−3.0
GDP.pdf (November 2026 rescoring)Professional › General knowledge work53.450.4−3.0
ChartographyPerception & media › Documents & charts73.571.0−2.5
FrontierMath Tiers 1-3 (v2) - PrivateMath91.293.7+2.5
Vals Science IndexScience › Multidisciplinary62.665.1+2.5
RiemannBench (November 2026 rescoring)Math84.081.6−2.4
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering75.878.0+2.3
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents19.214.9−2.2
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following72.970.8−2.1
Terminal-Bench 4.0Software › Software engineering61.659.6−2.0
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work33.432.2−1.2
Mystery Game PuzzlesGames85.084.0−1.0
BioMysteryBench (Vals Terminus 2)Science › Life sciences80.079.3−0.7
GPQA DiamondScience › Multidisciplinary96.395.8−0.5
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge75.275.6+0.4
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.01.2+0.1
FrontierMath Tier 4 (v2) - PrivateMath97.697.60.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
Terminal-Bench Science 0.1 (Vals)Science › Multidisciplinary62.962.90.0
Data: Vals AI, Surge AI, Artificial Analysis and 9 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…