Atlas

Compare Models

Gemini 4 Hydrogen vs Muse Wonder

Gemini 4 Hydrogen leads by 2.3 capability points (≈95% interval −0.2 to +4.7): statistically tied · P(higher) ≈ 96%

Wins 56 of 76 shared benchmarks (18 lost, 2 tied)

Capability vs. relative cost
Gemini 4 HydrogenMuse WonderClosed weightsOpen weights

Benchmark wins56 – 18 · 76 shared, 2 tied

Comparing matching setups only · suites counted once
← Gemini 4 HydrogenMuse Wonder →
Data: Artificial Analysis, Vals AI, Surge AI and 18 othersatlas.village.dev

Margins

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

EBR-bench v4 (card ban, single agent)Games70.531.4−39.0
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents13.50.6−25.2
HiL-Bench (Human-in-Loop Benchmark)Tool & computer use › General agents55.379.7+24.3
SWE-Marathon v1.2Software › Software engineering58.138.1−20.0
TaxCalcBench TY25Professional › Finance, accounting & tax44.024.0−20.0
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge58.839.0−19.8
Humanity's Sixth SensePerception & media › Spatial & visual reasoning63.443.7−19.7
Terminal-Bench-Science 0.1Science › Multidisciplinary41.421.9−19.5
GDP.xlsxProfessional › General knowledge work40.921.8−19.1
FrontierMath Tier 4 (v2) - PrivateMath87.870.7−17.1
KradleBenchGames43.927.6−16.3
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents29.3-3.0−16.1
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering54.038.4−15.7
sudo L7 (October 2026 rescoring)Software › Software engineering31.716.1−15.6
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations44.629.5−15.1
FrontierCode 2.0 - Pass RateSoftware › Software engineering43.228.8−14.4
Mystery Game PuzzlesGames65.051.0−14.0
FrontierCode 2.0 - ScoreSoftware › Software engineering38.925.3−13.5
Ramp SWE-Bench v2Software › Software engineering71.658.6−13.0
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering82.769.8−12.9
HLAB-Vals-1.1Professional › Legal11.323.8+12.5
DAYJOB: Law (November 2026 release)Professional › Legal9.720.6+10.9
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge46.325.1−10.6
Artificial Analysis Engineering IndexScience › Engineering60.149.6−10.5
Artificial Analysis Legal IndexProfessional › Legal56.746.9−9.9
GBENCH One-shot Coding GScoreGames66.858.5−8.3
LMArena Agent Arena Steering BurdenTool & computer use › General agents17.41.5−8.0
Artificial Analysis Coding Agent Index v2.1Software › Software engineering52.344.5−7.8
LMArena Code Arena WebDevSoftware › Frontend & design17031648−7.3
Tax Agent BenchProfessional › Finance, accounting & tax78.471.2−7.3
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine24.031.2+7.2
CritPt v2 (agentic, avg@5)Science › Physical sciences78.070.9−7.1
Artificial Analysis Finance & Accounting IndexProfessional › Finance, accounting & tax58.151.1−6.9
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents15.72.5−6.6
Professional Reasoning Benchmark - FinanceProfessional › Finance, accounting & tax52.358.7+6.4
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations74.468.1−6.3
LiveBench 2.0 (2026-11-06 release)Composite indexes77.070.7−6.3
AA-Briefcase v1.1 EloProfessional › General knowledge work14971610+6.2
FrontierMath Tiers 1-3 (v2) - PrivateMath88.182.1−6.0
micro1 PharmaBenchProfessional › General knowledge work11.417.2+5.8
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following73.767.9−5.8
Artificial Analysis Economics IndexComposite indexes55.049.5−5.5
Professional Reasoning Benchmark - LegalProfessional › Legal53.158.4+5.3
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine27.222.2−5.0
Agents' Last Exam v1.1 (ALE v1.1)Professional › General knowledge work43.438.5−4.9
Vals Finance IndexProfessional › Finance, accounting & tax66.161.2−4.9
Vibe Code Bench v1.1Software › Frontend & design87.883.1−4.7
Artificial Analysis Healthcare & Medical IndexHealth & medicine53.348.7−4.6
Terminal-Bench 4.0Software › Software engineering54.049.5−4.5
Epoch Capabilities Index (ECI)Composite indexes167162−4.1
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-4.53.4+4.0
CursorBench 4.0Software › Software engineering54.350.4−3.9
GradeBenchProfessional › Education51.948.1−3.8
FORTRESS Average Risk Score (ARS)Safety & alignment17.513.8+3.7
LMArena Text Arena EloLanguage & writing › General assistance14901511+3.6
Vals Index v2.2Composite indexes68.564.9−3.6
Legal Research BenchProfessional › Legal53.850.5−3.4
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning57.554.2−3.3
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax75.872.5−3.3
Artificial Analysis Strategy & Ops IndexProfessional › General knowledge work57.754.5−3.3
SWE-Marathon v1.1Software › Software engineering49.452.5+3.1
DAYJOB: AccountingProfessional › Finance, accounting & tax33.630.8−2.8
BuseyBench SVGPerception & media › Generation6.96.6−2.6
DAYJOB: TaxProfessional › Finance, accounting & tax3.66.0+2.4
Artificial Analysis Intelligence Index v5.0.1Composite indexes41.243.6+2.3
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work32.830.6−2.2
CWE-bench v2 Pass@1Software › Cybersecurity68.166.1−2.0
Vals Legal IndexProfessional › Legal73.675.3+1.6
Bug Hunt BenchSoftware › Software engineering31.130.2−1.0
GDP.pdf (November 2026 rescoring)Professional › General knowledge work54.254.8+0.6
OTIS Mock AIME 2024-2025Math100.099.4−0.6
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling0.7-0.2−0.4
Finance Agent v2.1Professional › Finance, accounting & tax65.165.5+0.4
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax38.538.8+0.3
Dual-Arm Sim (task completion)Tool & computer use › Robotics0.00.00.0
DualEntry Accounting AI BenchmarkProfessional › Finance, accounting & tax83.283.20.0
Data: Artificial Analysis, Vals AI, Surge AI and 18 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…