> ## Documentation Index
> Fetch the complete documentation index at: https://docs.priorlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Thinking mode

> Inference-time compute scaling for TabPFN — better predictions by spending more time at fit.

<Note>
  Thinking mode is only available through the TabPFN API (via `tabpfn-client` or the REST endpoints). It is not part of the open-source `tabpfn` package.
</Note>

Thinking mode introduces test-time compute scaling to TabPFN. Instead of a single forward pass, Thinking mode for TabPFN-3-Plus applies additional inference-time computation to push prediction quality further.

TabPFN-3-Plus with thinking mode beats all non-TabPFN models by over 200 Elo on the standard TabArena benchmark, rising to 420 Elo on the largest data subset. It outperforms AutoGluon 1.5 Extreme in 82% of the cases, in less than a tenth of its runtime, without using LLMs, real data, internet search, or any other model besides TabPFN. Check out our [model reports](/models) for details.

Thinking mode builds on top of TabPFN-3-Plus with native text-feature support, so a single call can handle mixed numerical, categorical, and text columns.

<Warning>
  Thinking fits consume a **monthly quota** separate from prediction tokens (default: 20 per month). Exceeding the limit returns HTTP 429. See [API metering](/api-reference/metering) for current limits and how to request more.
</Warning>

***

## When to use it

| Situation                                                                  | Guidance                                                              |
| -------------------------------------------------------------------------- | --------------------------------------------------------------------- |
| High-ROI use cases where small accuracy gains matter (finance, healthcare) | Use thinking mode — the one-time fit cost fuels recurring predictions |
| Pipelines with highly unstable ground truth data                           | Leave thinking off; use default TabPFN fit                            |

***

## Quickstart

<Note>
  Thinking mode is available only in the API with TabPFN-3-Plus. Local execution is not supported.
</Note>

<Tabs>
  <Tab title="Classification">
    ```python theme={null}
    from tabpfn_client import TabPFNClassifier

    clf = TabPFNClassifier(
        thinking_mode=True,
        thinking_effort="high",
        thinking_metric="accuracy",
    )
    clf.fit(X_train, y_train)
    preds = clf.predict(X_test)
    probs = clf.predict_proba(X_test)
    ```
  </Tab>

  <Tab title="Regression">
    ```python theme={null}
    from tabpfn_client import TabPFNRegressor

    reg = TabPFNRegressor(
        thinking_mode=True,
        thinking_effort="high",
        thinking_metric="rmse",
    )
    reg.fit(X_train, y_train)
    preds = reg.predict(X_test)
    ```
  </Tab>
</Tabs>

***

## Choosing effort and metric

`thinking_effort` controls how much compute is spent during fitting. `thinking_metric` sets the target.

**Effort levels:**

| Use case                                             | Recommended effort                                            |
| ---------------------------------------------------- | ------------------------------------------------------------- |
| You need maximum accuracy and can trade off fit time | Enable Thinking mode and set `thinking_effort="high"`         |
| You want a balance between quality and speed         | Start with default Thinking mode (`thinking_effort="medium"`) |

**Supported metrics:**

| Task           | Metrics                           |
| -------------- | --------------------------------- |
| Classification | `accuracy`, `log_loss`, `roc_auc` |
| Regression     | `rmse`, `mae`                     |

You can also set `thinking_timeout_s` to cap the wall-clock time spent on optimization.

***

## REST API

Call `POST /tabpfn/fit` with the thinking parameters in the JSON body:

```json theme={null}
{
  "train_set_upload_id": "<your-upload-id>",
  "task": "classification",
  "thinking_effort": "high",
  "thinking_timeout_s": 300,
  "thinking_effort_metric": "log_loss"
}
```

See the [API reference](/api-reference/getting-started) for the full endpoint documentation and upload flow.

***

## Parameters

| Parameter            | Type                   | Default | Description                                                                           |
| -------------------- | ---------------------- | ------- | ------------------------------------------------------------------------------------- |
| `thinking_mode`      | `bool`                 | `False` | Enable thinking mode. Setting `thinking_effort` also enables it.                      |
| `thinking_effort`    | `"medium"` or `"high"` | `None`  | Controls the effort & compute spent at fit time. Higher tends to give better results. |
| `thinking_timeout_s` | `float`                | `None`  | Wall-clock time budget in seconds                                                     |
| `thinking_metric`    | `str`                  | `None`  | Target metric to optimize (see supported metrics above)                               |

<Note>
  On the REST API, the metric parameter is called `thinking_effort_metric`.
</Note>

***

## Limits

Thinking fits have a **separate monthly quota** from prediction tokens. The default is **20 thinking fits per month**. When the quota is exhausted, `POST /tabpfn/fit` with thinking enabled returns HTTP **429**.

If you need higher limits, see [API metering](/api-reference/metering) for details or contact Prior Labs.

***

<CardGroup cols={2}>
  <Card title="TabPFN-3 changelog" icon="star" href="/changelog/tabpfn-3">
    Full release notes including thinking mode.
  </Card>

  <Card title="API metering" icon="gauge" href="/api-reference/metering">
    Token budgets, thinking fit limits, and usage tracking.
  </Card>

  <Card title="Classification" icon="tags" href="/capabilities/classification">
    Binary and multi-class classification guide.
  </Card>

  <Card title="Regression" icon="chart-line" href="/capabilities/regression">
    Point estimates, quantiles, and full distributions.
  </Card>
</CardGroup>