Atlas

Compare Models

Gemini 4 Argon vs Muse Ember

Gemini 4 Argon leads by 3.5 capability points (≈95% interval +1.0 to +5.9): a clear lead over Muse Ember · P(higher) ≈ 100%

Wins 44 of 53 shared benchmarks (9 lost)

Capability vs. relative cost
Gemini 4 ArgonMuse EmberClosed weightsOpen weights

Benchmark wins44 – 9 · 53 shared

Comparing matching setups only (3 benchmarks excluded) · suites counted once
← Gemini 4 ArgonMuse Ember →
Data: Artificial Analysis, Vals AI, Arena and 10 othersatlas.village.dev

Margins

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

Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge53.222.0−31.2
Terminal-Bench-Science 0.1Science › Multidisciplinary48.620.0−28.6
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering74.447.1−27.3
DAXBench v2Software › Data & ML88.362.8−25.5
GDP.xlsxProfessional › General knowledge work38.315.4−22.9
BuseyBench SVGPerception & media › Generation7.65.4−21.6
RiemannBench (November 2026 rescoring)Math74.453.6−20.8
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge46.17.7−19.2
SimpleBenchKnowledge & reasoning › General knowledge91.772.7−19.0
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents10.31.0−18.2
Terminal-Bench 4.0Software › Software engineering57.639.4−18.2
Artificial Analysis Intelligence Index v5.0.1Composite indexes50.436.0−14.4
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering55.140.9−14.1
SWE Atlas v1.1 - OverallSoftware › Software engineering70.156.0−14.1
GBENCH One-shot Coding GScoreGames69.756.4−13.3
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents23.8-2.7−13.2
DrugDiscoveryBenchScience › Life sciences65.452.4−13.0
HiL-Bench (Human-in-Loop Benchmark)Tool & computer use › General agents62.074.0+12.0
APEX-Agents 1.1Professional › General knowledge work87.475.7−11.7
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following70.358.8−11.5
CritPt v2 (agentic, avg@5)Science › Physical sciences72.962.3−10.6
ParseBench-2Perception & media › Documents & charts84.274.5−9.7
LMArena Code Arena WebDevSoftware › Frontend & design16811608−9.6
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations77.568.1−9.4
Tax Agent BenchProfessional › Finance, accounting & tax79.071.4−7.7
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents12.6-2.0−7.3
Artificial Analysis Coding Agent Index v2.0Software › Software engineering53.245.9−7.3
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations36.629.8−6.8
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine23.617.2−6.4
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax30.524.2−6.3
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents-6.06.6+6.3
Professional Reasoning Benchmark - LegalProfessional › Legal49.455.6+6.2
Professional Reasoning Benchmark - FinanceProfessional › Finance, accounting & tax49.956.0+6.1
Artificial Analysis Coding Agent Index v2.1Software › Software engineering45.539.7−5.8
Legal Research BenchProfessional › Legal54.849.0−5.8
Vals Finance IndexProfessional › Finance, accounting & tax66.060.9−5.1
DeepFinanceProfessional › Finance, accounting & tax41.436.5−4.9
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations58.153.3−4.8
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal27.422.7−4.7
Artificial Analysis Progress Index v1Composite indexes342329−4.6
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine13.918.3+4.4
GDP.pdf (November 2026 rescoring)Professional › General knowledge work53.649.2−4.4
Finance Agent v2.1Professional › Finance, accounting & tax67.763.8−3.8
LMArena Agent Arena Steering BurdenTool & computer use › General agents12.24.8−3.7
APEX-SWESoftware › Software engineering49.252.9+3.7
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work25.622.8−2.8
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax74.872.2−2.7
AA-Briefcase v1.1 EloProfessional › General knowledge work14851526+2.3
HLAB-Vals-1.1Professional › Legal27.125.0−2.1
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering25.727.2+1.5
Vals Legal IndexProfessional › Legal76.875.7−1.1
Ramp SWE-Bench v2 (Bash harness)Software › Software engineering54.854.4−0.4
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling0.80.9+0.1
Data: Artificial Analysis, Vals AI, Arena and 10 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…