Prior Labs

TabPFN-3.5-Thinking spends extra compute at fit time to get better at the metric you care about. In our new cookbook, that's worth €28,900 more in campaign value than XGBoost.

The setup: a bank ranks 200,000 customers by how likely they are to make a transaction, and only contacts the top of the list. Whose list finds more of the right people?

→ Thinking ranks the right customers higher than XGBoost and standard TabPFN-3.5
→ In its top 1%, more than 95 in 100 make a transaction
→ Contacting the top 10% earns about 8% more than with XGBoost

Simulated figures (€100 per transaction, €2 per contact), so plug in your own numbers.

Try it out: docs.priorlabs.ai/cookbook/predi…

New TabPFN-3.5-Thinking is live on our API.

The wait is over: run it on your own data today, in the playground or via the API.
Built for high-stakes predictions where greater accuracy is worth the investment:

→ 95% win rate against the previous #1 model on large, real-world data
→ The best accuracy to fitting time trade-off, with 32% less prediction time than before
→ Learns from up to 1M rows, 5x more than before
→ Best on grouped and temporal data, where patterns shift per customer, product and season

From power price forecasting to demand planning, customers are already putting it to work:
"𝘛𝘢𝘣𝘗𝘍𝘕 𝘛𝘩𝘪𝘯𝘬𝘪𝘯𝘨 𝘥𝘦𝘭𝘪𝘷𝘦𝘳𝘴 𝘣𝘦𝘵𝘵𝘦𝘳 𝘤𝘢𝘭𝘪𝘣𝘳𝘢𝘵𝘦𝘥 𝘲𝘶𝘢𝘯𝘵𝘪𝘭𝘦 𝘧𝘰𝘳𝘦𝘤𝘢𝘴𝘵𝘴 𝘵𝘩𝘢𝘯 𝘢𝘯𝘺 𝘮𝘦𝘵𝘩𝘰𝘥 𝘸𝘦 𝘩𝘢𝘷𝘦 𝘶𝘴𝘦𝘥, 𝘢𝘯𝘥 𝘪𝘵 𝘪𝘴 𝘯𝘰𝘸 𝘢 𝘬𝘦𝘺 𝘱𝘢𝘳𝘵 𝘰𝘧 𝘩𝘰𝘸 𝘸𝘦 𝘵𝘳𝘢𝘥𝘦."
- Christian Schütte, Head of Algo Trading and Quant, BD Energy

📣 Less than a week to go for our TabPFN-Rel live session

If you're interested in relational machine learning and how we're bringing tabular foundation models like TabPFN-3.5 to the field, this one is not to miss.

We'll cover everything from RelArena (the new benchmark for relational ML) to TabPFN-Rel, our new harness for TFMs, brining them relational ML capabilities. Expect hands on examples on how to use the harness, and more news on what's next for us at Prior Labs!

📅 November 4th, 6PM CET
🔗 Register: app.livestorm.co/prior-labs/mac…

Think more. Predict better. The new TabPFN-3.5-Thinking sets the SOTA bar.

How do you benchmark TabPFN against your current model?

New episode in our TabPFN Academy series: Philipp Singer (tag) walks through how he benchmarks TabPFN-3.5 against XGBoost.

Benchmarking is key if you want to evaluate TabPFN on your current baseline, so the tutorial and the notebook are yours: run it on your own data and see it for yourself.

youtube.com/watch?v=A7_rBO…