Atlas

Compare Models

Inkling 2 Small vs Muse Spark 1.3

Muse Spark 1.3 leads by 1.0 capability points (≈95% interval −1.0 to +3.1): statistically tied · P(higher) ≈ 84%

Wins 35 of 52 shared benchmarks (14 lost, 3 tied)

Capability vs. relative cost
Inkling 2 SmallMuse 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.

Inkling 2 SmallvsMuse Spark 1.3capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins14 – 35 · 52 shared, 3 tied

Comparing matching setups only (48 benchmarks excluded) · suites counted once
← Inkling 2 SmallMuse Spark 1.3 →

Inkling 2 SmallvsMuse Spark 1.3wins by categoryData Nov 15, 2026Data: Vals AI, Artificial Analysis, Surge AI and 12 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 Inkling 2 Small.

BrokenArXiv 08/2026 (Native Harnesses)Math63.120.2−42.9
Terminal-Bench 4.0 (Artificial Analysis)Code13.133.3+20.2
Legal Research Bench ImmigrationLegal27.345.5+18.2
Legal Research Bench FamilyLegal22.740.9+18.2
HANDBOOK.md Agents v1.1Professional Work17.833.9+16.1
AA-AnalystAgent (pass^5)Professional Work51.337.5−13.7
DAYJOB: Finance (October 2026 rescoring)Finance & Accounting14.327.0+12.7
CritPt v2 (agentic, avg@5)Science55.768.0+12.3
Terminal-Bench 4.0Code12.624.7+12.1
Legal Research Bench Business/CommercialLegal46.558.1+11.6
Legal Research Bench Civil LitigationLegal51.963.0+11.1
MysteryMechanismScience25.236.0+10.8
Legal Research Bench Constitutional/Civil RightsLegal39.550.0+10.5
Legal Research Bench CriminalLegal36.847.4+10.5
EnterpriseBench: CoreCraftProfessional Work71.861.9−9.9
APEX-Agents 1.1Professional Work60.970.4+9.5
ChartographyMultimodal36.727.6−9.1
ArXivMath 08/2026 (Native Harnesses)Math36.845.6+8.8
Tax Agent BenchFinance & Accounting65.873.4+7.6
Terminal-Bench 2.1Code81.388.8+7.5
SWE-Marathon v1.1Code38.846.3+7.5
SciCode 1.0.1 (Artificial Analysis)Code52.359.7+7.4
ComplexConstraintsChat & Writing44.551.9+7.4
DeepFinanceFinance & Accounting29.837.0+7.2
Terminal-Bench Science 0.1 (Vals)Science2.910.0+7.1
JobBench MainProfessional Work70.564.9−5.6
Design Arena Overall Code CategoriesCode13811348−5.5
Terminal-Bench-Science 0.1Science5.711.0+5.2
Humanity's Last Exam DiamondGeneral QA28.733.7+5.0
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine20.615.6−5.0
Vibe Code Bench v1.1Code81.585.9+4.3
DeepSWE v1.1Code72.167.8−4.3
Humanity's Sixth SenseMultimodal33.337.4+4.1
AA-Omniscience v1.1 IndexIndexes16.123.8+3.9
Legal Research Bench HealthLegal70.073.3+3.3
AA-Briefcase v1.1 EloAgents15271581+3.0
APEX-Accounting 1.1Finance & Accounting62.064.9+2.9
FrontierCode 2.0 (Artificial Analysis)Code25.423.1−2.3
GDPval-AA v2.1Professional Work16301681+2.2
ProgramBenchCode0.52.5+2.0
SRE Bench — Fully SolvedCode1.93.8+1.9
GPQA DiamondGeneral QA93.091.3−1.7
GDP.pdf (Artificial Analysis, All-pass)Professional Work25.226.6+1.4
Arena Code WebDev — ReactCode16591669+1.3
sudo L7 (October 2026 rescoring)Code14.112.8−1.3
Legal Research Bench Weighted PassLegal87.986.8−1.1
Arena Code WebDev — HTMLCode16601653−0.9
AA-3D v1.0 EloCode10271032+0.6
DeepSearchQA (F1 score)Search90.990.3−0.6
CyberBench v1.1 OverallCybersecurity72.772.70.0
Legal Research Bench Administrative/RegulatoryLegal60.660.60.0
SWE-bench Multimodal (Vals v1, 480 Tasks)Code43.143.10.0

Inkling 2 SmallvsMuse Spark 1.3normalised score marginsData Nov 15, 2026Data: Vals AI, Artificial Analysis, Surge AI and 12 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…