---
modelA: "Google TabFM"
modelB: "Kumo Tabular Small"
canonicalSlug: "tabfm-vs-kumo"
winsA: 11
winsB: 6
ties: 4
winRateA: "52.4%"
winRateB: "28.6%"
eloA: "1106 (±97)"
eloB: "1042 (±51)"
avgLatencyA: "742.64"
avgLatencyB: "14.52"
canonicalUrl: "https://carlahq.eu/datasets/comparison/tabfm-vs-kumo"
---

# TabFM vs Kumo Tabular Small Tabular Benchmark Comparison

> Empirical head-to-head evaluation across Carla's open-source tabular foundation model benchmark suite.

## Head-to-Head Scoreboard
| Metric | Google TabFM | Kumo Tabular Small | Comparison Result |
| :--- | :--- | :--- | :--- |
| **Head-to-Head Wins** | **11** (52.4%) | 6 (28.6%) | TabFM leads (4 ties) |
| **Bradley-Terry Elo Rating** | **1106 (±97)** | 1042 (±51) | Global tournament rating across repeated splits |
| **Improvability Regret** | 0.88% | 1.23% | Normalized average error against dataset optimal ceilings |
| **Average Inference Latency** | 742.64 | 14.52 | Measured per-split forward-pass timing |
| **Browser Execution** | Server-side only (Requires Python / JAX / PyTorch environment) | Native WebGPU & WASM Browser Runtime (via ONNX Runtime) | — |
| **Open Source License** | Open Source (Apache 2.0) | Open Source (Apache 2.0) | — |

## Architectural Specifications
| Dimension | Google TabFM | Kumo Tabular Small |
| :--- | :--- | :--- |
| **Organization** | Google Research | Kumo AI / NVIDIA |
| **Architecture** | Columnar Embedding & Self-Supervised In-Context Transformer | In-Context Tabular Foundation Model with Cross-Attention |
| **Pretraining Prior** | Generative synthetic distributions and masked column modeling | Relational & Structured Data Priors (SDM Recipe) |
| **Context Window** | Columnar token budget per batch | Configurable context rows with tiled row & ICL inference |
| **Execution Engine** | Server-side only (Requires Python / JAX / PyTorch environment) | Native WebGPU & WASM Browser Runtime (via ONNX Runtime) |
| **Repository** | [TabFM Code](https://github.com/google-research/tabfm) | [Kumo Tabular Small Code](https://huggingface.co/nvidia/Kumo-Tabular) |
| **Paper / ArXiv** | [TabFM Paper](https://github.com/google-research/tabfm) | [Kumo Tabular Small Paper](https://arxiv.org/abs/2502.05324) |

## Dataset-by-Dataset Benchmark Results
| Dataset | Business Domain | Task Type | Metric | TabFM Score | Kumo Tabular Small Score | Delta (%) | Winner |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| [Abalone](/datasets/undefined.md) | Agriculture, Forestry & Fishing | `regression` | undefined | 2.02 | 2.02 | +0.0% | **Kumo Tabular Small** |
| [Adult](/datasets/undefined.md) | Economics & Public Policy | `binary_classification` | undefined | 0.0677 | 0.0677 | +0.0% | **Tie** |
| [Airfoil Self Noise](/datasets/undefined.md) | UCI | `regression` | undefined | 1.11 | 1.02 | +0.1% | **Kumo Tabular Small** |
| [Amazon Employee Access](/datasets/undefined.md) | Information Technology & Enterprise Security | `binary_classification` | undefined | 0.1273 | 0.1392 | -0.0% | **TabFM** |
| [Bank Marketing](/datasets/undefined.md) | Finance & Banking | `binary_classification` | undefined | 0.1840 | 0.1880 | -0.0% | **TabFM** |
| [Blood Transfusion Service Center](/datasets/undefined.md) | Healthcare & Biomedicine | `binary_classification` | undefined | 0.2445 | 0.2452 | -0.0% | **TabFM** |
| [Breast W](/datasets/undefined.md) | Healthcare & Life Sciences | `binary_classification` | undefined | 0.0055 | 0.0055 | +0.0% | **Tie** |
| [Car](/datasets/undefined.md) | Automotive & Fleet Management | `multiclass_classification` | undefined | 0.0108 | 0.0253 | -0.0% | **TabFM** |
| [Compas Two Years](/datasets/undefined.md) | Legal & Public Safety | `binary_classification` | undefined | 0.2619 | 0.2617 | +0.0% | **Tie** |
| [Credit G](/datasets/undefined.md) | Finance & Banking | `binary_classification` | undefined | 0.1938 | 0.1956 | -0.0% | **TabFM** |
| [Diabetes](/datasets/undefined.md) | Healthcare & Biomedicine | `binary_classification` | undefined | 0.1623 | 0.1621 | +0.0% | **Kumo Tabular Small** |
| [Employee Salaries](/datasets/undefined.md) | Human Resources & Workforce Analytics | `regression` | undefined | 10,435 | 9,029 | +1405.5% | **Kumo Tabular Small** |
| [Fitness Club](/datasets/undefined.md) | Fitness, Sports & Recreation | `binary_classification` | undefined | 0.1795 | 0.1792 | +0.0% | **Kumo Tabular Small** |
| [House Sales](/datasets/undefined.md) | Real Estate & Property Valuation | `regression` | undefined | 98,887 | 98,512 | +374.3% | **Kumo Tabular Small** |
| [Houses](/datasets/undefined.md) | Real Estate & Urban Planning | `regression` | undefined | 37,481 | 37,670 | -189.2% | **TabFM** |
| [Monks Problems 2](/datasets/undefined.md) | Data Science & Artificial Intelligence | `binary_classification` | undefined | 0.0000 | 0.0000 | 0.0% | **Tie** |
| [Phoneme](/datasets/undefined.md) | Speech Processing & Acoustic Engineering | `binary_classification` | undefined | 0.0202 | 0.0246 | -0.0% | **TabFM** |
| [Spambase](/datasets/undefined.md) | Cybersecurity & IT Infrastructure | `binary_classification` | undefined | 0.0066 | 0.0073 | -0.0% | **TabFM** |
| [Telco Customer Churn](/datasets/undefined.md) | Telecommunications & Subscription Services | `binary_classification` | undefined | 0.1474 | 0.1595 | -0.0% | **TabFM** |
| [Titanic](/datasets/undefined.md) | Maritime Safety & Actuarial Risk Analysis | `binary_classification` | undefined | 0.1166 | 0.1276 | -0.0% | **TabFM** |
| [Vehicle](/datasets/undefined.md) | Automotive Engineering & Computer Vision | `multiclass_classification` | undefined | 0.1986 | 0.2244 | -0.0% | **TabFM** |

## Canonical Resources
- **HTML Benchmark Page:** [https://carlahq.eu/datasets/comparison/tabfm-vs-kumo](https://carlahq.eu/datasets/comparison/tabfm-vs-kumo)
- **Carla Tabular Suite:** [https://carlahq.eu/datasets/](https://carlahq.eu/datasets/)
