Atlas

Compare Models

Gemini 4 Hydrogen vs GPT-6.2 Astra

GPT-6.2 Astra leads by 2.8 capability points (≈95% interval +0.1 to +5.6): a clear lead over Gemini 4 Hydrogen · P(higher) ≈ 98%

Wins 52 of 74 shared benchmarks (19 lost, 3 tied)

Capability vs. relative cost
Gemini 4 HydrogenGPT-6.2 AstraClosed weightsOpen weights

Benchmark wins19 – 52 · 74 shared, 3 tied

Comparing matching setups only (31 benchmarks excluded) · suites counted once
← Gemini 4 HydrogenGPT-6.2 Astra →
Data: Artificial Analysis, Epoch AI, Surge AI and 17 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.2 Astra minus Gemini 4 Hydrogen.

KradleBenchGames43.978.3+34.4
SpireClimbGames7.741.3+33.7
DAYJOB: Law (November 2026 release)Professional › Legal9.741.4+31.7
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning57.589.2+31.7
Mystery Game PuzzlesGames65.095.0+30.0
Terminal-Bench-Science 0.1Science › Multidisciplinary49.578.6+29.0
EBR-bench v4 (card ban, single agent)Games70.592.4+21.9
Artificial Analysis Intelligence Index v5.0.1Composite indexes41.262.1+20.9
Tax Agent BenchProfessional › Finance, accounting & tax78.457.9−20.6
Chess Puzzles (task version 2.0)Games60.078.0+18.0
SimpleBenchKnowledge & reasoning › General knowledge91.777.3−14.4
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge58.872.2+13.4
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine27.239.4+12.2
FrontierMath Tier 4 (v2) - PrivateMath87.8100.0+12.2
GDP.xlsxProfessional › General knowledge work40.929.7−11.2
Legal Research BenchProfessional › Legal53.842.8−11.1
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering53.564.1+10.6
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine24.034.4+10.4
Professional Reasoning Benchmark - LegalProfessional › Legal53.143.1−10.0
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering39.049.0+10.0
Terminal-Bench 4.0Software › Software engineering54.063.1+9.1
Vals Finance IndexProfessional › Finance, accounting & tax66.157.3−8.8
DAYJOB: AccountingProfessional › Finance, accounting & tax33.642.0+8.4
LiveBench 2.0 (2026-11-06 release)Composite indexes77.084.9+7.9
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge70.978.5+7.6
MultiChallengeLanguage & writing › Conversation80.673.2−7.4
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax38.545.8+7.3
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work71.678.3+6.7
TxBench Preclinical Pharmacology V1 Clean RerunsScience › Life sciences61.368.0+6.7
sudo L7 (October 2026 rescoring)Software › Software engineering31.738.3+6.6
Artificial Analysis Engineering IndexScience › Engineering60.166.6+6.5
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-4.57.7+6.1
Artificial Analysis Economics IndexComposite indexes55.061.1+6.0
DAYJOB: TaxProfessional › Finance, accounting & tax3.69.2+5.6
Artificial Analysis Strategy & Ops IndexProfessional › General knowledge work57.763.3+5.5
DualEntry Accounting AI BenchmarkProfessional › Finance, accounting & tax83.278.2−5.0
GDP.pdf (November 2026 rescoring)Professional › General knowledge work54.249.2−5.0
HLAB-Vals-1.1Professional › Legal11.316.3+5.0
Epoch Capabilities Index (ECI)Composite indexes167173+4.8
BigFinanceBench private holdout (final-answer accuracy)Professional › Finance, accounting & tax49.454.0+4.6
FrontierMath Tiers 1-3 (v2) - PrivateMath88.192.6+4.6
Finance Agent v2.1Professional › Finance, accounting & tax65.160.7−4.4
Ramp SWE-Bench v2 (Bash harness)Software › Software engineering70.875.0+4.2
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following73.769.6−4.1
Vals Legal IndexProfessional › Legal73.669.7−3.9
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents29.337.0+3.9
Artificial Analysis Healthcare & Medical IndexHealth & medicine53.357.0+3.7
DrugDiscoveryBenchScience › Life sciences69.165.5−3.6
Chartography (November 2026 rescoring)Perception & media › Documents & charts87.290.7+3.5
ARC-AGI-2Knowledge & reasoning › Abstract reasoning92.595.8+3.3
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents13.515.2+3.3
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work32.829.8−3.0
Artificial Analysis Finance & Accounting IndexProfessional › Finance, accounting & tax58.160.9+2.9
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations74.477.2+2.8
CritPt v2 (agentic, avg@5)Science › Physical sciences80.683.1+2.6
Dual-Arm Sim (task completion)Tool & computer use › Robotics0.02.5+2.5
Gray Swan IPI Attack Success (k=15)Safety & alignment0.02.4−2.4
Vibe Code Bench v1.1Software › Frontend & design87.889.9+2.1
Artificial Analysis Legal IndexProfessional › Legal56.758.7+2.0
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge46.350.0+1.9
LMArena Agent Arena Steering BurdenTool & computer use › General agents17.414.0−1.7
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents15.712.4−1.6
AA-Briefcase v1.1 EloProfessional › General knowledge work15691594+1.4
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations51.152.4+1.2
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations44.645.8+1.2
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering82.783.6+0.9
Vals Index v2.2Composite indexes68.567.7−0.8
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling0.71.2+0.2
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax75.876.0+0.2
Humanity's Sixth SensePerception & media › Spatial & visual reasoning63.463.2−0.1
GPQA DiamondScience › Multidisciplinary96.896.8+0.1
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning98.598.50.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
TaxCalcBench TY25Professional › Finance, accounting & tax44.044.00.0
Data: Artificial Analysis, Epoch AI, Surge AI and 17 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…