Atlas

Compare Models

Gemini 4 Hydrogen vs Muse Spark 1.3

Gemini 4 Hydrogen leads by 4.7 capability points (≈95% interval +2.3 to +7.0): a clear lead over Muse Spark 1.3 · P(higher) ≈ 100%

Wins 50 of 60 shared benchmarks (9 lost, 1 tied)

Capability vs. relative cost
Gemini 4 HydrogenMuse Spark 1.3Closed weightsOpen weights

Benchmark wins50 – 9 · 60 shared, 1 tied

Comparing matching setups only (2 benchmarks excluded) · suites counted once
← Gemini 4 HydrogenMuse Spark 1.3 →
Data: Artificial Analysis, Surge AI, Arena and 17 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Muse Spark 1.3 minus Gemini 4 Hydrogen.

Terminal-Bench-Science 0.1Science › Multidisciplinary49.53.8−45.7
Terminal-Bench 4.0Software › Software engineering54.010.6−43.4
FrontierMath Tier 4 (v2) - PrivateMath87.846.3−41.5
Mystery Game PuzzlesGames65.025.0−40.0
ChartographyPerception & media › Documents & charts66.627.6−39.0
TaxCalcBench TY25Professional › Finance, accounting & tax44.06.0−38.0
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering53.516.7−36.9
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations51.120.7−30.4
FrontierSWE V2Software › Software engineering53.125.8−27.3
Humanity's Sixth SensePerception & media › Spatial & visual reasoning63.437.4−26.0
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge58.833.7−25.1
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents13.51.1−24.2
Chess Puzzles (task version 2.0)Games60.037.0−23.0
CursorBench 4.1Software › Software engineering61.441.9−19.5
sudo L7 (October 2026 rescoring)Software › Software engineering31.712.8−18.9
GDP.xlsxProfessional › General knowledge work40.922.6−18.3
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations74.457.9−16.6
Gray Swan IPI Attack Success (k=15)Safety & alignment0.015.9−15.9
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering39.023.1−15.8
DAYJOB: AccountingProfessional › Finance, accounting & tax33.618.0−15.6
Artificial Analysis Coding Agent Index v2.1Software › Software engineering52.336.8−15.5
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents29.30.1−14.6
FrontierMath Tiers 1-3 (v2) - PrivateMath88.174.4−13.7
GBENCH One-shot Coding GScoreGames66.853.7−13.1
CursorBench 4.0Software › Software engineering54.341.6−12.7
CritPt v2 (agentic, avg@5)Science › Physical sciences80.668.0−12.6
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine27.215.6−11.7
HiL-Bench (Human-in-Loop Benchmark)Tool & computer use › General agents55.366.9+11.6
BuseyBench SVGPerception & media › Generation6.95.7−11.5
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax38.527.0−11.5
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge46.323.8−11.2
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations44.633.9−10.7
GradeBenchProfessional › Education51.941.4−10.5
SimpleBenchKnowledge & reasoning › General knowledge91.781.8−9.9
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following73.764.8−8.9
Professional Reasoning Benchmark - LegalProfessional › Legal53.161.6+8.5
LiveBench 2.0 (2026-11-06 release)Composite indexes77.068.6−8.4
LMArena Agent Arena Steering BurdenTool & computer use › General agents17.40.8−8.3
Professional Reasoning Benchmark - FinanceProfessional › Finance, accounting & tax52.359.5+7.3
DrugDiscoveryBenchScience › Life sciences69.162.2−6.9
Epoch Capabilities Index (ECI)Composite indexes167158−6.9
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents15.72.5−6.6
LMArena Code Arena WebDevSoftware › Frontend & design17031654−6.5
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work32.826.6−6.2
Tax Agent BenchProfessional › Finance, accounting & tax78.473.4−5.1
Vibe Code Bench v1.1Software › Frontend & design87.882.9−4.9
AA-Briefcase v1.1 EloProfessional › General knowledge work15691480−4.9
Artificial Analysis Intelligence Index v5.0.1Composite indexes41.236.6−4.6
SWE-Marathon v1.1Software › Software engineering49.446.3−3.1
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-4.51.2+2.8
micro1 PharmaBenchProfessional › General knowledge work11.413.9+2.5
DAYJOB: Law (November 2026 release)Professional › Legal9.711.8+2.1
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine24.022.0−2.0
Legal Research BenchProfessional › Legal53.855.3+1.4
LMArena Text Arena EloLanguage & writing › General assistance14901498+1.4
GDP.pdf (November 2026 rescoring)Professional › General knowledge work54.253.0−1.2
OTIS Mock AIME 2024-2025Math100.099.2−0.8
Bug Hunt BenchSoftware › Software engineering31.130.7−0.4
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling0.71.0+0.2
DAYJOB: TaxProfessional › Finance, accounting & tax3.63.60.0
Data: Artificial Analysis, Surge AI, Arena 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…