Atlas

Compare Models

GPT-5.5 vs Inkling 2 Small

Inkling 2 Small leads by 0.1 capability points (≈95% interval −1.9 to +2.1): statistically tied · P(higher) ≈ 55%

Wins 31 of 47 shared benchmarks (14 lost, 2 tied)

Capability vs. relative cost
GPT-5.5Inkling 2 SmallClosed weightsOpen weights

Benchmark wins14 – 31 · 47 shared, 2 tied

Comparing matching setups only (7 benchmarks excluded) · suites counted once
← GPT-5.5Inkling 2 Small →
Data: Vals AI, Artificial Analysis, Thinking Machines Lab and 13 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Inkling 2 Small minus GPT-5.5.

JobBench MainProfessional › General knowledge work38.370.5+32.2
AA-Briefcase v1.1 EloProfessional › General knowledge work11351527+21.5
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations51.371.8+20.5
Apex (MathArena)Math80.263.5−16.7
GDPval-AA v2.1Professional › General knowledge work13571630+12.0
Vibe Code Bench v1.1Software › Frontend & design69.881.5+11.7
LMArena Text Arena EloLanguage & writing › General assistance14761412−11.2
Legal Research Bench Constitutional/Civil RightsProfessional › Legal28.939.5+10.5
MCP Atlas April 2026 (all 1,000 tasks)Tool & computer use › General agents75.385.4+10.1
BabyVision with PythonPerception & media › Spatial & visual reasoning83.693.6+10.0
APEX-Agents 1.1Professional › General knowledge work70.360.9−9.4
Legal Research Bench Business/CommercialProfessional › Legal37.246.5+9.3
Legal Research Bench FamilyProfessional › Legal13.622.7+9.1
Legal Research Bench ImmigrationProfessional › Legal36.427.3−9.1
PerceptionBenchPerception & media › General visual understanding55.864.7+8.9
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine28.920.6−8.3
DeepSWE v1.1Software › Software engineering64.472.1+7.7
Legal Research Bench Administrative/RegulatoryProfessional › Legal53.060.6+7.6
Legal Research Bench Civil LitigationProfessional › Legal44.451.9+7.4
BrowseCompTool & computer use › Search & research84.490.2+5.8
APEX-Accounting 1.1Professional › Finance, accounting & tax56.362.0+5.7
ChartographyPerception & media › Documents & charts31.036.7+5.7
Tax Agent BenchProfessional › Finance, accounting & tax60.565.8+5.3
MedCodeHealth & medicine49.144.1−5.0
ComplexConstraintsLanguage & writing › Instruction following49.544.5−5.0
Legal Research Bench Weighted PassProfessional › Legal84.087.9+3.9
SciCode 1.0.1 (Artificial Analysis)Science › Multidisciplinary56.152.3−3.8
SWE-bench MultilingualSoftware › Software engineering77.781.2+3.5
SWE-bench VerifiedSoftware › Software engineering78.882.2+3.4
ARC-AGI-2Knowledge & reasoning › Abstract reasoning85.081.7−3.3
Legal Research Bench HealthProfessional › Legal66.770.0+3.3
MedScribeHealth & medicine86.990.2+3.3
DeepSearchQA (F1 score)Tool & computer use › Search & research87.890.9+3.1
Terminal-Bench 2.1Software › Software engineering84.381.3−3.0
HMMT Feb 2026Math98.595.8−2.7
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work22.625.2+2.6
ARC-AGI-3 (Semi-Private)Games0.42.6+2.1
SRE Bench — Fully SolvedSoftware › Cybersecurity3.81.9−1.9
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering14.613.1−1.5
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning95.096.5+1.5
CharXiv ReasoningPerception & media › Documents & charts84.185.6+1.5
AA-AnalystAgent (pass^5)Professional › General knowledge work50.051.3+1.2
Artificial Analysis Progress Index v1Composite indexes320318−0.9
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge27.928.7+0.8
EnterpriseOps-Gym-AATool & computer use › Customer service & operations46.646.8+0.2
Legal Research Bench CriminalProfessional › Legal36.836.80.0
ProgramBenchSoftware › Software engineering0.50.50.0
Data: Vals AI, Artificial Analysis, Thinking Machines Lab and 13 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…