# Tabular Foundation Model Leaderboard & Comparisons

> Empirical tournament evaluation using Bradley-Terry Elo ratings with 95% bootstrap confidence intervals across repeated cross-validation splits.

## Global Elo Leaderboard

| Rank | Model | Organization | Bradley-Terry Elo | 95% CI | Improvability | Latency | In-Browser WebGPU | License |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| 1 | [**Google TabFM**](/datasets/models/tabfm.md) | Google Research | **1111** | ±89 | 0.85% | 742.64s | Server only | Open Source (Apache 2.0) |
| 2 | [**EXAONE Tabular**](/datasets/models/exaone-tabular.md) | LG AI Research | **1021** | ±77 | 0.85% | 158.67s | Server only | Open Source (Apache 2.0) |
| 3 | [**TabPFN v3**](/datasets/models/tabpfn.md) | Prior Labs | **936** | ±58 | 1.69% | 12.52s | PyTorch only | Research & Non-Commercial |
| 4 | [**TabICLv2**](/datasets/models/tabiclv2.md) | Inria SODA | **932** | ±61 | 2.11% | 4.26s | WebGPU (WASM) | Open Source (BSD-3-Clause) |

## Pairwise Model Comparisons

- [TabICLv2 vs TabPFN v3](/datasets/comparison/tabiclv2-vs-tabpfn.md): Detailed head-to-head win rates, Elo deltas, and per-dataset scorecards.
- [TabICLv2 vs EXAONE Tabular](/datasets/comparison/tabiclv2-vs-exaone-tabular.md): Detailed head-to-head win rates, Elo deltas, and per-dataset scorecards.
- [TabICLv2 vs Google TabFM](/datasets/comparison/tabiclv2-vs-google-tabfm.md): Detailed head-to-head win rates, Elo deltas, and per-dataset scorecards.
- [TabPFN v3 vs EXAONE Tabular](/datasets/comparison/tabpfn-vs-exaone-tabular.md): Detailed head-to-head win rates, Elo deltas, and per-dataset scorecards.
- [TabPFN v3 vs Google TabFM](/datasets/comparison/tabpfn-vs-google-tabfm.md): Detailed head-to-head win rates, Elo deltas, and per-dataset scorecards.
- [EXAONE Tabular vs Google TabFM](/datasets/comparison/exaone-tabular-vs-google-tabfm.md): Detailed head-to-head win rates, Elo deltas, and per-dataset scorecards.