Sold by: Prior Labs
Deployed on AWS
TabPFN-3-Plus by Prior Labs is the latest generation Tabular Foundation Model. A single forward pass tops the public TabArena benchmark for classification and regression, scales to 1M training rows (at 200 features) or 100k rows (at 2000 features), and runs up to 20x faster than TabPFN-2.5. TabPFN-3-Plus also natively handles text features: string-valued columns are accepted directly, without requiring upstream featurization, and are encoded jointly with numeric and categorical features inside the model. Released under the TABPFN-3.0 License v1.0 for research and internal evaluation.
Overview
TabPFN-3-Plus by Prior Labs is the latest generation of the leading Tabular Foundation Model. Pretrained exclusively on synthetic data from a structural-causal-model prior, it produces calibrated predictions for classification and regression in a single forward pass with no model selection and no hyperparameter tuning. On the public TabArena benchmark, a forward pass tops every prior tabular foundation model and Pareto-dominates the speed-versus-performance frontier. Validated scale: up to 1M training rows at 200 features, 100k at 2000 features, or 1k at 20000 features. A redesigned architecture with per-row in-context learning, two-stage row compression, KV cache, and row chunking runs up to 20x faster than TabPFN-2.5 and fits 1M rows on a single H100 GPU. Native many-class classification, mixed feature types, missing values, and outliers are handled directly. TabPFN-3-Plus also includes native text-feature support: string-valued columns (free-text fields such as product names, claim descriptions, or customer reviews) are accepted directly and encoded jointly with the numeric and categorical columns inside the model, so cross-feature interactions between text and structured columns are learned end-to-end rather than imposed by a fixed encoder. Released under the TABPFN-3.0 License v1.0 for research and internal evaluation; commercial licenses via Prior Labs.
Highlights
- Tops the public TabArena benchmark in a single forward pass. State-of-the-art classification and regression with no model selection and no hyperparameter tuning.
- Scales beyond any prior TabPFN: validated up to 1M rows at 200 features, 100k at 2000, or 1k at 20000. Up to 20x faster than TabPFN-2.5 and fits 1M rows on a single H100 GPU.
- Native handling of mixed feature types: numerical, categorical, free-text (string-valued) columns, missing values, and outliers. No bespoke preprocessing or upstream featurization required.
Details
Sold by
Delivery method
Latest version
Deployed on AWS
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Financing for AWS Marketplace purchases
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
Pricing
This product is available free of charge. Free subscriptions have no end date and may be canceled any time.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Vendor refund policy
Using TabPFN-3 on Amazon SageMaker is free of charge for non-commercial use. Because the model itself does not incur any licensing or usage fees, no refunds are provided for any costs incurred while using this product, including but not limited to charges for AWS compute instances, storage, networking, or any other AWS infrastructure used to run or host the model.
If you have questions about this policy or need assistance, you can contact Prior Labs at: hello@priorlabs.ai
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA) .
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
Delivery details
Amazon SageMaker model
An Amazon SageMaker model package is a pre-trained machine learning model ready to use without additional training. Use the model package to create a model on Amazon SageMaker for real-time inference or batch processing. Amazon SageMaker is a fully managed platform for building, training, and deploying machine learning models at scale.
Deploy the model on Amazon SageMaker AI using the following options:
Version release notes
Initial release of TabPFN-3 on Amazon SageMaker. Includes full v3-schema documentation and v2.5 -> v3 migration notes in UsageInstructions.
Additional details
Inputs
- Summary
TabPFN-3 accepts inference requests in two content types:
- application/json -- inline JSON-encoded request body.
- multipart/form-data -- with X_train, y_train, X_test as separate Parquet or CSV file parts plus a JSON config part.
Sync (real-time) invocations are capped by SageMaker at 6 MB payload and 60 s processing. Async invocations support up to 1 GB payload and 60 min processing, which is the recommended deployment for any non-trivial tabular dataset.
The canonical v3 request body shape is flat: top-level X_train / y_train / X_test, plus a top-level task_config object containing task, tabpfn_config (model parameters), and predict_params. For backward compatibility the server also accepts the legacy v2.5 layout where task, tabpfn_config (or its alias model_params), and predict_params are siblings of the dataset fields at the top level. The v2.5 nested data wrapper ({"data": {"x_train": ..., "y_train": ..., "x_test": ...}}) is NOT accepted -- see the migration notes in UsageInstructions.
- Limitations for input type
- Validated TabPFN-3 scale envelope: up to 1,000,000 training rows at 200 features; up to 100,000 rows at 2,000 features; up to 1,000 rows at 20,000 features. Predictions outside this envelope are still computed when `ignore_pretraining_limits` is true (the default), but accuracy is not guaranteed. `n_estimators` is capped at 8. SageMaker payload caps: 6 MB sync / 1 GB async per request; sync processing capped at 60 s. Real-time inference is the only supported sync mode; batch transform is supported on a subset of supported instance types (see RecommendedInstanceTypes).
{"task_config": {"task": "classification", "tabpfn_config": {"n_estimators": 4, "fit_mode": "fit_preprocessors"}, "predict_params": {"output_type": "probas"}}, "X_train": [[1.2, 3.4], [5.6, 7.8], [2.1, 4.3], [6.0, 1.5]], "y_train": [0, 1, 0, 1], "X_test": [[1.0, 4.0], [5.0, 2.0], [3.0, 3.0]]}
{"task_config": {"task": "classification", "tabpfn_config": {"n_estimators": 4, "fit_mode": "fit_preprocessors"}, "predict_params": {"output_type": "probas"}}, "X_train": [[1.2, 3.4], [5.6, 7.8], [2.1, 4.3], [6.0, 1.5]], "y_train": [0, 1, 0, 1], "X_test": [[1.0, 4.0], [5.0, 2.0], [3.0, 3.0]]}
Input data descriptions
The following table describes supported input data fields for real-time inference and batch transform.
Field name | Description | Constraints | Required |
|---|---|---|---|
task | Inference task type. Required; accepted at top level (legacy v2.5 shape) or nested under `task_config.task` (canonical v3 shape). | AllowedValues: classification, regression. | Yes |
n_estimators | Number of TabPFN forward passes that are averaged for the final prediction. Default is the underlying tabpfn library default. | MinValue: 1. MaxValue: 8. | No |
categorical_features_indices | Indices of columns to treat as categorical. If omitted or null, the server infers categorical columns from the data. | - | No |
softmax_temperature | Temperature for the softmax over class logits (classification) or the predictive distribution (regression). | MinValue: greater than 0. Default: 0.9. | No |
average_before_softmax | Whether to average ensemble logits before applying softmax (instead of averaging post-softmax probabilities). Only meaningful when n_estimators > 1. | Boolean. Default: false. | No |
ignore_pretraining_limits | If true, the server does not reject inputs that exceed TabPFN-3's validated row / feature / cell-budget envelope. Outside-envelope inputs may run but with degraded accuracy. | Boolean. Default: true (CHANGED from TabPFN-2.5 where the default was false). | No |
inference_precision | Numeric precision used for the forward pass. "auto" selects per-instance defaults. | Default: auto. | No |
fit_mode | Controls how the server caches between the fit-time forward pass and any subsequent predicts. `fit_preprocessors` (default) runs everything per request. `fit_with_cache` caches the fit and returns a model_id which can be reused on later requests via `context.model_id` to skip the fit step. | AllowedValues: fit_preprocessors, fit_with_cache. Default: fit_preprocessors. | No |
memory_saving_mode | If true, reduces GPU/CPU memory pressure during inference. Trades a small amount of throughput for the ability to handle larger contexts. | Boolean or null. Default: null (auto-decided by the server based on input size and instance type). | No |
random_state | Seeds preprocessing-side randomness for reproducibility across runs. | Integer or null. Default: null. | No |
Resources
Vendor resources
Support
Vendor support
For general support or license inquiries reach out to us via hello@priorlabs.ai
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
Similar products
TabPFN-2.5 is a Tabular Foundation Model (TFM) used to make predictions on structured data without hyperparameter tuning and with minimal preprocessing. It is a pretrained transformer that provides instant predictions for classification, regression, and time-series tasks across messy tabular data (numeric, text, categorical, missing values, outliers, uninformative features). TabPFN-2.5 handles datasets up to 50,000 samples and 2,000 features and outperforms tree-based models and multi-hour tuned ensembles. This TabPFN-2.5 model package is free to use under the non-commercial conditions as specified in the model license.
Customer reviews
No customer reviews yet
Be the first to review this product . We've partnered with PeerSpot to gather customer feedback. You can share your experience by writing or recording a review, or scheduling a call with a PeerSpot analyst.