Atlas

Compare Models

Gemini 4 Carbon vs Muse Spark 1.3

Gemini 4 Carbon leads by 7.0 capability points (≈95% interval +4.4 to +9.5): a clear lead over Muse Spark 1.3 · P(higher) ≈ 100%

Wins 51 of 54 shared benchmarks (3 lost)

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

Each line runs through one model’s priced reasoning settings, cheapest to most expensive; grey points are other models’ priced settings in the same range. Results whose reasoning effort the source didn’t report are left out wherever a model has named effort levels. Capability and relative cost are the published estimates (relative cost is cost per task as a multiple of GPT-5 (high)); hover a setting, or use the arrow keys, to set it beside the other model’s same setting.

Gemini 4 CarbonvsMuse Spark 1.3capability vs. relative costData Nov 17, 2026Village LLM Atlasatlas.village.dev

Benchmark wins51 – 3 · 54 shared

Comparing matching setups only (4 benchmarks excluded) · suites counted once
← Gemini 4 CarbonMuse Spark 1.3 →

Gemini 4 CarbonvsMuse Spark 1.3wins by categoryData Nov 17, 2026Data: Artificial Analysis, Arena, Epoch AI and 13 othersVillage LLM Atlasatlas.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 Carbon.

BrokenArXiv 08/2026 (Native Harnesses)Math88.120.2−67.9
RiemannBenchMath80.028.0−52.0
Terminal-Bench-Science 0.1Science62.411.0−51.4
FrontierMath Tier 4 (v2) - PrivateMath95.146.3−48.8
Mystery Game PuzzlesGames73.025.0−48.0
ArXivMath 08/2026 (Native Harnesses)Math93.045.6−47.4
ChartographyMultimodal72.527.6−44.9
FrontierSWE V2 Worst@5Code56.212.6−43.6
Chess Puzzles (task version 2.0)Games78.037.0−41.0
Terminal-Bench 4.0Code63.124.7−38.4
Humanity's Last Exam DiamondGeneral QA69.133.7−35.4
AutomationBench 1.0.6 (Zapier)Agents53.020.7−32.3
Terminal-Bench 4.0 (Artificial Analysis)Code62.633.3−29.3
LMArena Agent Arena Net Improvement (Steering Burden)Agents16.41.2−29.0
FrontierSWE V2 Best@5Code66.740.9−25.8
Bug Hunt BenchCode54.630.7−23.9
Artificial Analysis Intelligence Index v5.0.1Indexes60.036.6−23.4
BuseyBench SVGMultimodal7.95.7−21.5
Artificial Analysis Intelligence Index v5.0Indexes57.736.4−21.3
LMArena Agent Arena Praise vs ComplaintAgents40.30.1−20.1
APEX-Agents 1.1Professional Work90.470.4−20.0
AutomationBench-AA 1.0.6Agents77.857.9−20.0
CursorBench 4.0Code60.441.6−18.8
Artificial Analysis Coding Agent Index v2.1Indexes55.536.8−18.7
FrontierCode 2.0 (Artificial Analysis)Code40.623.1−17.4
FrontierMath Tiers 1-3 (v2) - PrivateMath90.574.4−16.1
CWE-bench v1 Programmatic Pass@1Cybersecurity71.055.0−16.0
GBENCH One-shot Coding GScoreGames69.053.7−15.3
GradeBenchProfessional Work56.641.4−15.2
LMArena Code Arena WebDevCode17681655−15.0
AA-Omniscience v1.1 IndexIndexes53.523.8−14.8
CritPt v2 (agentic, avg@5)Science82.368.0−14.3
Artificial Analysis Coding Agent Index v2.0Indexes59.445.7−13.7
Artificial Analysis Progress Index v1Indexes351324−9.2
Epoch Capabilities Index (ECI)Indexes170158−9.2
LMArena Agent Arena Steering BurdenAgents18.11.0−8.6
DAXBench v2Code93.685.1−8.5
Vals RSI Index v1.1 (four tasks)Agents27.719.6−8.1
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal23.015.0−8.0
LMArena Agent Arena Confirmed SuccessAgents17.92.7−7.6
APEX-Accounting 1.1Finance & Accounting70.764.9−5.8
LMArena Text Arena EloChat & Writing15261497−5.2
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine19.415.6−3.9
LMArena Agent Arena Bash ReliabilityAgents-5.91.6+3.8
AA-LCR v1.1Search79.383.0+3.7
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.752.2−3.5
Vibe Code Bench v1.1Code89.185.9−3.2
Tax Agent BenchFinance & Accounting76.473.4−3.0
GDP.pdf (Artificial Analysis, All-pass)Professional Work29.626.6−3.0
GDPval-AA v2.1Professional Work17331681−2.3
AA-Briefcase v1.1 EloAgents16111581−1.6
OTIS Mock AIME 2024-2025Math100.099.2−0.8
Legal Research BenchLegal55.855.3−0.5
LMArena Agent Arena Tool HallucinationAgents0.60.8+0.1

Gemini 4 CarbonvsMuse Spark 1.3normalised score marginsData Nov 17, 2026Data: Artificial Analysis, Arena, Epoch AI and 13 othersVillage LLM Atlasatlas.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…