--- title: NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency... description: NVIDIA released Kumo Tabular, an open foundation model for tabular classification and regression available on Hugging Face and GitHub under the OpenMDW-1.1... canonical: https://daily.dev/posts/nvidia-kumo-tabular-sets-a-new-accuracy-efficiency-frontier-for-tabular-prediction-avcjvvwsc twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction | daily.dev og:description: NVIDIA released Kumo Tabular, an open foundation model for tabular classification and regression available on Hugging Face and GitHub under the OpenMDW-1.1... og:url: https://daily.dev/posts/nvidia-kumo-tabular-sets-a-new-accuracy-efficiency-frontier-for-tabular-prediction-avcjvvwsc og:image: https://api.daily.dev/og/posts/avcjvVWsc.png og:image:alt: NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction **[Hugging Face](https://daily.dev/sources/huggingface)** · 8 min read · 0 upvotes · 0 comments ## Summary NVIDIA released Kumo Tabular, an open foundation model for tabular classification and regression available on Hugging Face and GitHub under the OpenMDW-1.1 license. It predicts labels for new rows in a single forward pass with no training or feature engineering, using in-context learning inspired by TabICL and TabPFN. Pretrained entirely on synthetic tables generated via structural causal models, it comes in three sizes (28M to 215M parameters) and ranks first on TabArena, BeyondArena, TALENT, and ScoringBench benchmarks while running significantly faster than competing models like LimiX-2. Limitations include support for only numerical and categorical columns natively and degraded accuracy on out-of-distribution data. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is NVIDIA Kumo Tabular and how does it make predictions on tabular data? Kumo Tabular is an open foundation model for tabular classification and regression that predicts labels for new rows in a single forward pass, without training, tuning, or feature engineering. Given a table of labeled rows as context, it uses in-context learning, similar to TabICL and TabPFN, applying column, row, and in-context attention. It comes in three sizes from 28M to 215M parameters, released under the OpenMDW-1.1 license. _daily.dev surfaces releases like this for engineers evaluating foundation models against gradient-boosted trees for tabular tasks._ ### How does Kumo Tabular's accuracy and speed compare to LimiX-2 on TabArena? Kumo Tabular ranks first overall on the TabArena leaderboard with an ELO of 1950, while running 26 times faster than LimiX-2 under a uniform single RTX 6000 Pro GPU evaluation setup. It establishes a new state-of-the-art on the accuracy-efficiency Pareto front across all three model sizes, outperforming tuned gradient-boosted trees, AutoGluon, and other tabular foundation models. _track benchmark comparisons like this on daily.dev when choosing between tabular foundation models and boosted trees._ ### What data was NVIDIA Kumo Tabular trained on and what are its limitations? Kumo Tabular was pretrained entirely on artificial tables sampled from structural causal models, generating millions of synthetic tables with varied causal graphs, missingness patterns, and outliers rather than real-world data. It handles only numerical and categorical columns natively, supports up to 10 classes per forward pass, and accuracy may degrade on tables far outside training ranges or with distribution shift between context and query rows. _developers weighing synthetic-data-trained models for production can follow model details like this on daily.dev._ ## Similar posts on daily.dev - [Kumo’s new foundation model replaces months of data science engineering with plain-English queries](https://daily.dev/posts/kumo-s-new-foundation-model-replaces-months-of-data-science-engineering-with-plain-english-queries-xrcvjrwyq) · The New Stack · 0 upvotes · 0 comments - [Tabular LLMs: An Introduction to the Foundation Models That Predict Your Spreadsheet](https://daily.dev/posts/tabular-llms-an-introduction-to-the-foundation-models-that-predict-your-spreadsheet-dkoxe4tky) · Towards Data Science · 0 upvotes · 0 comments - [H2O.ai launches tabH2O foundation model for tabular data](https://daily.dev/posts/h2o-ai-launches-tabh2o-foundation-model-for-tabular-data-hwllsnijd) · The Next Web · 0 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#machine-learning](https://daily.dev/tags/machine-learning), [#nvidia](https://daily.dev/tags/nvidia) [View this post on daily.dev](https://daily.dev/posts/nvidia-kumo-tabular-sets-a-new-accuracy-efficiency-frontier-for-tabular-prediction-avcjvvwsc) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction","url":"https://daily.dev/posts/nvidia-kumo-tabular-sets-a-new-accuracy-efficiency-frontier-for-tabular-prediction-avcjvvwsc","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/nvidia-kumo-tabular-sets-a-new-accuracy-efficiency-frontier-for-tabular-prediction-avcjvvwsc"},"datePublished":"2026-09-29T15:33:55.821Z","dateModified":"2026-09-29T15:34:17.977Z","description":"NVIDIA released Kumo Tabular, an open foundation model for tabular classification and regression available on Hugging Face and GitHub under the OpenMDW-1.1...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/040ba2c88d27b382012ae05d4a74041c?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/040ba2c88d27b382012ae05d4a74041c?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hugging Face","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hugging Face","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/f1f55c67d81a4330acf5b90b26b0c8e1","url":"https://daily.dev/sources/huggingface"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/nvidia-kumo-tabular-sets-a-new-accuracy-efficiency-frontier-for-tabular-prediction-avcjvvwsc","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,machine-learning,nvidia","timeRequired":"PT8M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hugging Face","item":"https://daily.dev/sources/huggingface"},{"@type":"ListItem","position":3,"name":"NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/nvidia-kumo-tabular-sets-a-new-accuracy-efficiency-frontier-for-tabular-prediction-avcjvvwsc#faq","mainEntity":[{"@type":"Question","name":"What is NVIDIA Kumo Tabular and how does it make predictions on tabular data?","acceptedAnswer":{"@type":"Answer","text":"Kumo Tabular is an open foundation model for tabular classification and regression that predicts labels for new rows in a single forward pass, without training, tuning, or feature engineering. Given a table of labeled rows as context, it uses in-context learning, similar to TabICL and TabPFN, applying column, row, and in-context attention. It comes in three sizes from 28M to 215M parameters, released under the OpenMDW-1.1 license. daily.dev surfaces releases like this for engineers evaluating foundation models against gradient-boosted trees for tabular tasks."}},{"@type":"Question","name":"How does Kumo Tabular's accuracy and speed compare to LimiX-2 on TabArena?","acceptedAnswer":{"@type":"Answer","text":"Kumo Tabular ranks first overall on the TabArena leaderboard with an ELO of 1950, while running 26 times faster than LimiX-2 under a uniform single RTX 6000 Pro GPU evaluation setup. It establishes a new state-of-the-art on the accuracy-efficiency Pareto front across all three model sizes, outperforming tuned gradient-boosted trees, AutoGluon, and other tabular foundation models. track benchmark comparisons like this on daily.dev when choosing between tabular foundation models and boosted trees."}},{"@type":"Question","name":"What data was NVIDIA Kumo Tabular trained on and what are its limitations?","acceptedAnswer":{"@type":"Answer","text":"Kumo Tabular was pretrained entirely on artificial tables sampled from structural causal models, generating millions of synthetic tables with varied causal graphs, missingness patterns, and outliers rather than real-world data. It handles only numerical and categorical columns natively, supports up to 10 classes per forward pass, and accuracy may degrade on tables far outside training ranges or with distribution shift between context and query rows. developers weighing synthetic-data-trained models for production can follow model details like this on daily.dev."}}]} ```