Atlas

Compare Models

Gemini 4 Hydrogen vs Grok 4.6

Gemini 4 Hydrogen leads by 6.0 capability points (≈95% interval +3.7 to +8.3): a clear lead over Grok 4.6 · P(higher) ≈ 100%

Wins 48 of 52 shared benchmarks (4 lost)

Capability vs. relative cost
Gemini 4 HydrogenGrok 4.6Closed weightsOpen weights

Benchmark wins48 – 4 · 52 shared

Comparing matching setups only · suites counted once
← Gemini 4 HydrogenGrok 4.6 →
Data: Artificial Analysis, Epoch AI, Arena and 15 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Grok 4.6 minus Gemini 4 Hydrogen.

FrontierMath Tier 4 (v2) - PrivateMath87.831.7−56.1
Gray Swan IPI Attack Success (k=15)Safety & alignment0.051.8−51.8
Terminal-Bench-Science 0.1Science › Multidisciplinary49.52.9−46.7
ChartographyPerception & media › Documents & charts66.620.5−46.1
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge58.821.5−37.3
Terminal-Bench 4.0Software › Software engineering54.017.2−36.9
TxBench-Antibody-Development V1 (60 evaluations)Science › Life sciences60.626.1−34.4
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering54.021.2−32.8
Mystery Game PuzzlesGames65.034.0−31.0
FrontierSWE V2Software › Software engineering53.125.3−27.8
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents13.5-0.6−27.5
Chess Puzzles (task version 2.0)Games60.033.0−27.0
ARC-AGI-2Knowledge & reasoning › Abstract reasoning92.567.1−25.4
BuseyBench SVGPerception & media › Generation6.94.6−22.5
FrontierMath Tiers 1-3 (v2) - PrivateMath88.166.0−22.1
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge70.949.3−21.6
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering39.019.8−19.2
CritPt v2 (agentic, avg@5)Science › Physical sciences80.662.3−18.3
SWE-Marathon v1.1Software › Software engineering49.431.9−17.5
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax38.521.0−17.5
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning57.540.0−17.5
CursorBench 4.1Software › Software engineering61.444.3−17.1
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents29.3-3.5−16.4
SimpleBenchKnowledge & reasoning › General knowledge91.775.9−15.8
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work32.817.8−15.0
GBENCH One-shot Coding GScoreGames66.852.3−14.5
LiveBench 2.0 (2026-11-06 release)Composite indexes77.063.5−13.5
CursorBench 4.0Software › Software engineering54.341.4−12.9
Vibe Code Bench v1.1Software › Frontend & design87.876.2−11.5
LMArena Code Arena WebDevSoftware › Frontend & design17031618−11.3
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning98.587.5−11.0
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge46.324.9−10.7
Artificial Analysis Intelligence Index v5.0.1Composite indexes41.231.0−10.2
LMArena Agent Arena Steering BurdenTool & computer use › General agents17.4-2.5−10.0
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine24.014.8−9.2
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents15.7-2.5−9.1
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine27.218.3−8.9
TxBench Oligo Discovery V1 (113 evaluations)Science › Life sciences54.346.0−8.3
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations74.467.0−7.4
Epoch Capabilities Index (ECI)Composite indexes167158−7.2
Tax Agent BenchProfessional › Finance, accounting & tax78.471.4−7.1
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-4.59.1+6.8
Legal Research BenchProfessional › Legal53.848.1−5.8
LMArena Text Arena EloLanguage & writing › General assistance14901457−5.8
TxBench Antibody Discovery V1 (100 evaluations)Science › Life sciences57.753.0−4.7
Bug Hunt BenchSoftware › Software engineering31.127.3−3.8
GPQA DiamondScience › Multidisciplinary96.894.0−2.8
AA-Briefcase v1.1 EloProfessional › General knowledge work15691540−1.6
micro1 PharmaBenchProfessional › General knowledge work11.412.5+1.1
OTIS Mock AIME 2024-2025Math100.099.2−0.8
TxBench Preclinical Pharmacology V1 Clean RerunsScience › Life sciences61.361.7+0.3
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling0.71.2+0.2
Data: Artificial Analysis, Epoch AI, Arena and 15 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…