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.
Model Specifications & Design Trade-offs
Comparing transformer architecture, Bayesian priors, sequence context windows, and browser execution runtime.
TabICLv2
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.
Kumo Tabular Small
Kumo AI and NVIDIA's in-context tabular foundation model, pre-trained on diverse structured datasets for zero-shot classification and quantile regression.
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.
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).
| Dataset | Domain | Task | Rows | 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 & Fishing | REG | 4,177 | 3×3 | RMSE | 2.03 | 2.02 | +0.01 | Kumo Tabular Small | 1.07s | 1.75s |
| Adult ↗ | Economics & Public Policy | BIN | 48,842 | 3×3 | 1 - AUC | 0.0793 | 0.0677 | +0.0116 | Kumo Tabular Small | 23.84s | 96.87s |
| Airfoil Self Noise ↗ | UCI | REG | 1,503 | 10×3 | RMSE | 1.12 | 1.02 | +0.10 | Kumo Tabular Small | 0.65s | 0.73s |
| Amazon Employee Access ↗ | Information Technology & Enterprise Security | BIN | 32,769 | 3×3 | 1 - AUC | 0.1483 | 0.1392 | +0.0091 | Kumo Tabular Small | 11.93s | 48.40s |
| Bank Marketing ↗ | Finance & Banking | BIN | 45,211 | 3×3 | 1 - AUC | 0.1970 | 0.1880 | +0.0090 | Kumo Tabular Small | 21.49s | 85.00s |
| Blood Transfusion Service Center ↗ | Healthcare & Biomedicine | BIN | 748 | 20×3 | 1 - AUC | 0.2446 | 0.2452 | -0.0006 | TabICLv2 | 0.61s | 0.50s |
| Breast W ↗ | Healthcare & Life Sciences | BIN | 699 | 20×3 | 1 - AUC | 0.0052 | 0.0055 | -0.0003 | TabICLv2 | 0.67s | 0.53s |
| Car ↗ | Automotive & Fleet Management | MULTI | 1,728 | 10×3 | Log-Loss | 0.0231 | 0.0253 | -0.0022 | TabICLv2 | 0.74s | 0.65s |
| Compas Two Years ↗ | Legal & Public Safety | BIN | 5,278 | 3×3 | 1 - AUC | 0.2701 | 0.2617 | +0.0084 | Kumo Tabular Small | 1.73s | 2.74s |
| Credit G ↗ | Finance & Banking | BIN | 1,000 | 10×3 | 1 - AUC | 0.2016 | 0.1956 | +0.0061 | Kumo Tabular Small | 0.89s | 0.77s |
| Diabetes ↗ | Healthcare & Biomedicine | BIN | 768 | 10×3 | 1 - AUC | 0.1632 | 0.1621 | +0.0011 | Kumo Tabular Small | 0.65s | 0.54s |
| Employee Salaries ↗ | Human Resources & Workforce Analytics | REG | 9,228 | 3×3 | RMSE | 8,126 | 9,029 | -903 | TabICLv2 | 2.13s | 5.53s |
| Fitness Club ↗ | Fitness, Sports & Recreation | BIN | 1,500 | 10×3 | 1 - AUC | 0.1795 | 0.1792 | +0.0002 | Kumo Tabular Small | 0.70s | 0.76s |
| House Sales ↗ | Real Estate & Property Valuation | REG | 21,613 | 3×3 | RMSE | 111,524 | 98,512 | +13,011 | Kumo Tabular Small | 7.52s | 26.72s |
| Houses ↗ | Real Estate & Urban Planning | REG | 20,640 | 3×3 | RMSE | 40,971 | 37,670 | +3,301 | Kumo Tabular Small | 5.04s | 19.70s |
| Monks Problems 2 ↗ | Data Science & Artificial Intelligence | BIN | 601 | 20×3 | 1 - AUC | 0.0000 | 0.0000 | ±0.00 | Tie | 0.60s | 0.50s |
| Phoneme ↗ | Speech Processing & Acoustic Engineering | BIN | 5,404 | 3×3 | 1 - AUC | 0.0278 | 0.0246 | +0.0033 | Kumo Tabular Small | 1.37s | 2.29s |
| Spambase ↗ | Cybersecurity & IT Infrastructure | BIN | 4,601 | 3×3 | 1 - AUC | 0.0075 | 0.0073 | +0.0002 | Kumo Tabular Small | 3.66s | 4.98s |
| Telco Customer Churn ↗ | Telecommunications & Subscription Services | BIN | 7,043 | 3×3 | 1 - AUC | 0.1488 | 0.1595 | -0.0107 | TabICLv2 | 2.63s | 4.67s |
| Titanic ↗ | Maritime Safety & Actuarial Risk Analysis | BIN | 1,309 | 10×3 | 1 - AUC | 0.1229 | 0.1276 | -0.0047 | TabICLv2 | 0.78s | 0.84s |
| Vehicle ↗ | Automotive Engineering & Computer Vision | MULTI | 846 | 10×3 | Log-Loss | 0.2401 | 0.2244 | +0.0157 | Kumo Tabular Small | 0.82s | 0.53s |
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: 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.