Atlas

Compare Models

Opus 4.8 vs Inkling 2 Small

Inkling 2 Small leads by 0.4 capability points (≈95% interval −1.6 to +2.4): statistically tied · P(higher) ≈ 65%

Wins 32 of 47 shared benchmarks (12 lost, 3 tied)

Capability vs. relative cost
Opus 4.8Inkling 2 SmallClosed 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.

Opus 4.8vsInkling 2 Smallcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins12 – 32 · 47 shared, 3 tied

Comparing matching setups only (8 benchmarks excluded) · suites counted once
← Opus 4.8Inkling 2 Small →

Opus 4.8vsInkling 2 Smallwins by categoryData Nov 15, 2026Data: Vals AI, Artificial Analysis, Thinking Machines Lab and 13 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Inkling 2 Small minus Opus 4.8.

JobBench MainProfessional Work48.470.5+22.1
EnterpriseBench: CoreCraftProfessional Work52.371.8+19.5
Apex (MathArena)Math81.363.5−17.7
DeepSWE v1.1Code54.472.1+17.7
PerceptionBenchMultimodal47.264.7+17.5
Legal Research Bench Administrative/RegulatoryLegal43.960.6+16.7
Legal Research Bench HealthLegal53.370.0+16.7
Arena Code WebDev — HTMLCode15421660+16.5
Arena Code WebDev — ReactCode15561659+13.8
BabyVision with PythonMultimodal81.293.6+12.4
AA-Briefcase v1.1 EloAgents13201527+11.3
LMArena Text Arena EloChat & Writing14741412−10.9
Terminal-Bench 4.0Code23.212.6−10.6
Humanity's Sixth SenseMultimodal23.533.3+9.8
ARC-AGI-2General QA72.181.7+9.6
ComplexConstraintsChat & Writing35.644.5+8.9
Legal Research Bench CriminalLegal45.636.8−8.8
Terminal-Bench 4.0 (Artificial Analysis)Code21.713.1−8.6
FrontierCode 2.0 - Pass RateCode17.125.5+8.4
GDPval-AA v2.1Professional Work14611630+7.4
FrontierCode 2.0 - ScoreCode15.322.6+7.3
Chartography (with tools)Multimodal75.082.1+7.1
AA-AnalystAgent (pass^5)Professional Work45.051.3+6.2
Humanity's Last Exam DiamondGeneral QA23.128.7+5.6
CharXiv ReasoningMultimodal80.585.6+5.1
Legal Research Bench Business/CommercialLegal41.946.5+4.7
Legal Research Bench FamilyLegal18.222.7+4.5
ARC-AGI-1 Semi-PrivateGeneral QA92.596.5+4.0
APEX-Agents 1.1Professional Work64.460.9−3.5
MCP Atlas April 2026 (all 1,000 tasks)Agents82.285.4+3.2
EnterpriseOps-Gym-AAAgents44.046.8+2.9
GDP.pdf (Artificial Analysis, All-pass)Professional Work22.825.2+2.4
SciCode 1.0.1 (Artificial Analysis)Code54.452.3−2.1
Legal Research Bench Weighted PassLegal86.187.9+1.8
ScreenSpot-ProMultimodal82.383.8+1.5
Terminal-Bench Science 0.1 (Vals)Science4.32.9−1.4
DeepSearchQA (F1 score)Search92.290.9−1.3
Vibe Code Bench v1.1Code82.781.5−1.2
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine19.420.6+1.1
PostTrainBench v1.2Indexes31.031.9+0.9
Tax Agent BenchFinance & Accounting64.965.8+0.8
Terminal-Bench 2.1Code82.081.3−0.7
ProgramBenchCode1.00.5−0.5
HMMT Feb 2026Math95.595.8+0.3
Legal Research Bench Civil LitigationLegal51.951.90.0
Legal Research Bench Constitutional/Civil RightsLegal39.539.50.0
Legal Research Bench ImmigrationLegal27.327.30.0

Opus 4.8vsInkling 2 Smallnormalised score marginsData Nov 15, 2026Data: Vals AI, Artificial Analysis, Thinking Machines Lab 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…