Prior Labs ships TabPFN-3.5, a tabular model trained on synthetic data
single source· 1 articles · confidence: low · first seen 2026-09-16 06:52 UTC
What this means for you
Nothing to act on yet. No licence, weights, pricing or availability details are given, and the comparison against the Otto entry has no published evaluation date or harness, so the claim cannot be checked. Note the name; wait for numbers you can reproduce.
Prior Labs has released TabPFN-3.5, a model for tabular data — the spreadsheet-shaped rows and columns most business records live in — pretrained only on synthetic data. The company says it outperforms the winning entry in the Otto Kaggle competition using default settings. What the release does not give: an evaluation date, a harness description, licence or pricing terms, model size, or any sign the result has been checked outside Prior Labs. The single press item carries the claim without figures.
Key facts
- ·Prior Labs released TabPFN-3.5, described as a tabular foundation model. source
- ·The model was pretrained only on synthetic data. source
- ·Prior Labs says it beats the winning Otto Kaggle solution using default settings. source
- ·No evaluation date, harness, or score is given for the Otto comparison. source
- ·The item was published on 16 September 2026. source
What the sources say
- MarkTechPost — Brief release note; no harness, evaluation date, score or licence given for the claimed competition result.
Sources
The original reporting. Follow these — they did the work.