Skip to main content
CARLA HQ
PAIRWISE BENCHMARKAdaptive Repeated 3-Fold21 DatasetsMarkdown ↗

TabICLv2 vs Kumo Tabular Small Benchmark

Empirical head-to-head evaluation of TabICLv2 (Inria SODA) versus Kumo Tabular Small (Kumo AI / NVIDIA) on identical, precomputed IID or non-IID splits across 21 diverse tabular datasets.

HEAD-TO-HEAD WINS?Direct pairwise dataset matchup wins where the model achieved lower out-of-sample error loss (1 - AUC for binary, Log-Loss for multiclass, RMSE for regression).
TabICLv26
:
14Kumo Tabular Small
(1 tie) across 21 benchmarks
TOURNAMENT ELO?Global Bradley-Terry Elo rating computed across all evaluation-split matchups with 1,000 bootstrap resamples. Higher is better (baseline 1000).
TabICLv2916 (±65)
vs
Kumo Tabular Small1042 (±55)
Higher is better • 95% Bootstrap CI
MACRO IMPROVABILITY?Normalized average error regret relative to empirical suite ceilings across all 21 datasets. Lower is better (0.0% is optimal ceiling).
TabICLv22.13%
vs
Kumo Tabular Small1.23%
Lower is better • Error regret vs ceiling
MEAN SPLIT LATENCY?Average execution latency in seconds per persisted IID or non-IID evaluation split across all benchmarks. Lower is faster.
TabICLv24.26s
vs
Kumo Tabular Small14.52s
Lower is faster • adaptive split protocol
ARCHITECTURAL BLUEPRINTS

Model Specifications & Design Trade-offs

Comparing transformer architecture, Bayesian priors, sequence context windows, and browser execution runtime.

Inria SODA

TabICLv2

GitHub / Paper ↗

Inria SODA's state-of-the-art tabular foundation model engineered for instantaneous in-context learning. TabICLv2 runs zero-shot inference without iterative gradient descent and is fully optimized for client-side WebGPU execution in Google Sheets via Carla HQ.

ArchitectureIn-Context Tabular Transformer with Prefix Attention & KV Caching
Prior / PretrainingSynthetic Copula-based Quantile Priors & Prior.py DAGs
Context Window1,000+ context rows with zero fine-tuning required
Runtime PlatformNative WebGPU Browser Runtime (Carla Engine / ONNX Runtime WASM)
LicenseOpen Source (BSD-3-Clause)
Kumo AI / NVIDIA

Kumo Tabular Small

GitHub / Paper ↗

Kumo AI and NVIDIA's in-context tabular foundation model, pre-trained on diverse structured datasets for zero-shot classification and quantile regression.

ArchitectureIn-Context Tabular Foundation Model with Cross-Attention
Prior / PretrainingRelational & Structured Data Priors (SDM Recipe)
Context WindowConfigurable context rows with tiled row & ICL inference
Runtime PlatformNative WebGPU & WASM Browser Runtime (via ONNX Runtime)
LicenseOpen Source (Apache 2.0)
DEEP DIVE REPORTON-DEVICE RUNTIME ANALYSIS

TabICLv2 vs TabPFN-3 vs TabFM vs EXAONE: Which Model Can Run in the Browser?

Evaluating model weight size (MB vs GB), VRAM requirements, Python server infrastructure overhead, licensing terms, and WebGPU client-side execution compatibility.

HEAD-TO-HEAD BREAKDOWN

21-Dataset Metric Evaluation Matrix

Empirical performance on deterministic, application-appropriate IID or non-IID splits: 1 - AUC for binary classification, Log-Loss for multiclass, and RMSE for regression (all metrics: lower is better).

DatasetDomainTaskRows
Splits?Repeats × folds. IID datasets use three folds with size-aware repeats, and the same row indices are used for every model.
Metric?Primary evaluation error loss: 1 - AUC for binary classification, Log-Loss for multiclass, RMSE for regression. All metrics: lower is better.
TabICLv2
Loss?TabICLv2 primary out-of-sample error loss (lower is better).
Kumo Tabular Small
Loss?Kumo Tabular Small primary out-of-sample error loss (lower is better).
Margin (Δ)?Error loss difference. Negative (green) indicates TabICLv2 lead; positive (red) indicates Kumo Tabular Small lead.
Winner?Model achieving lower out-of-sample error loss on this benchmark split.
TabICLv2
Latency?Mean inference runtime in seconds per evaluation split for TabICLv2.
Kumo Tabular Small
Latency?Mean inference runtime in seconds per evaluation split for Kumo Tabular Small.
Abalone ↗Agriculture, Forestry & FishingREG4,1773×3RMSE2.032.02+0.01Kumo Tabular Small1.07s1.75s
Adult ↗Economics & Public PolicyBIN48,8423×31 - AUC0.07930.0677+0.0116Kumo Tabular Small23.84s96.87s
Airfoil Self Noise ↗UCIREG1,50310×3RMSE1.121.02+0.10Kumo Tabular Small0.65s0.73s
Amazon Employee Access ↗Information Technology & Enterprise SecurityBIN32,7693×31 - AUC0.14830.1392+0.0091Kumo Tabular Small11.93s48.40s
Bank Marketing ↗Finance & BankingBIN45,2113×31 - AUC0.19700.1880+0.0090Kumo Tabular Small21.49s85.00s
Blood Transfusion Service Center ↗Healthcare & BiomedicineBIN74820×31 - AUC0.24460.2452-0.0006TabICLv20.61s0.50s
Breast W ↗Healthcare & Life SciencesBIN69920×31 - AUC0.00520.0055-0.0003TabICLv20.67s0.53s
Car ↗Automotive & Fleet ManagementMULTI1,72810×3Log-Loss0.02310.0253-0.0022TabICLv20.74s0.65s
Compas Two Years ↗Legal & Public SafetyBIN5,2783×31 - AUC0.27010.2617+0.0084Kumo Tabular Small1.73s2.74s
Credit G ↗Finance & BankingBIN1,00010×31 - AUC0.20160.1956+0.0061Kumo Tabular Small0.89s0.77s
Diabetes ↗Healthcare & BiomedicineBIN76810×31 - AUC0.16320.1621+0.0011Kumo Tabular Small0.65s0.54s
Employee Salaries ↗Human Resources & Workforce AnalyticsREG9,2283×3RMSE8,1269,029-903TabICLv22.13s5.53s
Fitness Club ↗Fitness, Sports & RecreationBIN1,50010×31 - AUC0.17950.1792+0.0002Kumo Tabular Small0.70s0.76s
House Sales ↗Real Estate & Property ValuationREG21,6133×3RMSE111,52498,512+13,011Kumo Tabular Small7.52s26.72s
Houses ↗Real Estate & Urban PlanningREG20,6403×3RMSE40,97137,670+3,301Kumo Tabular Small5.04s19.70s
Monks Problems 2 ↗Data Science & Artificial IntelligenceBIN60120×31 - AUC0.00000.0000±0.00Tie0.60s0.50s
Phoneme ↗Speech Processing & Acoustic EngineeringBIN5,4043×31 - AUC0.02780.0246+0.0033Kumo Tabular Small1.37s2.29s
Spambase ↗Cybersecurity & IT InfrastructureBIN4,6013×31 - AUC0.00750.0073+0.0002Kumo Tabular Small3.66s4.98s
Telco Customer Churn ↗Telecommunications & Subscription ServicesBIN7,0433×31 - AUC0.14880.1595-0.0107TabICLv22.63s4.67s
Titanic ↗Maritime Safety & Actuarial Risk AnalysisBIN1,30910×31 - AUC0.12290.1276-0.0047TabICLv20.78s0.84s
Vehicle ↗Automotive Engineering & Computer VisionMULTI84610×3Log-Loss0.24010.2244+0.0157Kumo Tabular Small0.82s0.53s
ZERO-SERVER CLIENT-SIDE EXECUTION

Evaluate Foundation Models In Your Spreadsheets

Carla HQ brings TabICLv2 foundation models directly into Google Sheets via WebGPU. Run zero-trust predictions on private data without API keys or cloud pipelines.

FREQUENTLY ASKED QUESTIONS

Frequently Asked Questions: TabICLv2 vs Kumo Tabular Small

Common architectural and benchmarking questions comparing TabICLv2 and Kumo Tabular Small.

How does TabICLv2 compare to Kumo Tabular Small on tabular benchmarks?

Across the Carla benchmark suite, TabICLv2 and Kumo Tabular Small are evaluated head-to-head on identical, precomputed IID or non-IID splits using standardized error losses (1 - AUC for binary classification, Log-Loss for multiclass, and RMSE for regression). IID datasets use adaptive repeated 3-fold evaluation to reduce variance. TabICLv2 enables 100% private in-browser WebGPU execution, whereas server-side models like Kumo Tabular Small require heavy PyTorch infrastructure and cloud GPU backend servers.

Can TabICLv2 or Kumo Tabular Small run in the browser without server infrastructure?

Only TabICLv2 (via Carla HQ and nanotabicl ONNX Runtime WASM/WebGPU) is engineered to execute entirely client-side inside web browsers like Google Chrome. Kumo Tabular Small requires full Python and PyTorch server infrastructure with high memory allocations.

What are the primary architectural differences between TabICLv2 and Kumo Tabular Small?

TabICLv2 uses in-context tabular transformer with prefix attention & kv caching, whereas Kumo Tabular Small leverages in-context tabular foundation model with cross-attention. TabICLv2 specifically utilizes prefix attention with KV cache reuse for real-time tabular evaluation in spreadsheets.