Sphinx AI

Have a lot of different notebooks you've been using with Sphinx? It's now much easier to keep track of all your Sphinx analyses, since the chat history is now global!

Sphinx can now search the internet for useful context, powered by the awesome tools of Parallel Web Systems (@p0)! Sphinx is especially adept at using this tool to search for open datasets and APIs to use in analyses, such as stock prices, weather data, and sports results.

Thank you to @Adyen and @huggingface for the DABstep benchmark. This work helps define what great looks like and raises the bar for everyone. While Sphinx is moving away from DABstep as a primary benchmark after a test-set leak, we’ll still use it in a limited capacity. [1/2]

Sphinx also is now able to perform Jupyter-aware file search. This is important because if your Jupyter server is hosted on a remote machine (like via Colab), your kernel will only be able to access that remote machine's file system. Sphinx understands this.

0.8 also lets you pick from one of three different Engines, each offering a different personality and tradeoff of speed and quality. Engines are carefully curated and we set a high bar for inclusion in Sphinx, so that users have only the best.

Plan Mode is a massive improvement to Sphinx's ability to tackle complicated and underspecified analyses. With Plan Mode selected, Sphinx will generate a detailed strategy for how to approach each problem, considering details like feature selection, confounds, and data cleaning.

No compute? No problem! You can now use Sphinx with Google Colab's VS Code extension to build complex models and analyses in your IDE, all totally free and with no need for local resources! See Sphinx build a CNN model on MNIST on a Colab GPU:

Even over a wide range of random scatterplots, frontier models only have a crude understanding of data! This is why Sphinx’s representation learning is critical to ensuring that data science gets done correctly.

Here’s why we think this happens – even if an agent naively looks at code and plots, AI is not good at building data intuition in the same way as a human. For example, we gave GPT-5 the following plot and asked it for the correlation, it estimated 0% (correct answer: -48%)

Unfortunately, house prices don’t start to fall after 3,200 sq ft! This model doesn't really make sense. When we toss this task to Sphinx, it excises the bad data and arrives at a much more reasonable solution