🕳️ Static IP proxies

You can now easily connect to databases and other private resources without exposing them to the public internet. Modal Proxy lets you route outbound traffic through static IPs exclusive to your workspace.

💬 Slack integration

Need to receive real-time alerts for your Modal apps? You can now do this with our native Slack integration.

🌈 Live usage dashboard

You can now track resource usage across all the apps in your workspace. To see this view, go to Live Usage in your dashboard and then click on the expand button next to Active Resources.

🍁 GPU fallbacks

You can now specify GPU fallbacks for your functions, which will increase the pool of resources your function can be scheduled on.

👩‍💻 Client Updates

Run pip install --upgrade modal to get the latest updates. Here are some of the highlights:

• CLI: modal shell now takes a container ID, allowing you to shell into a running container.
• Functions: Memory snapshotting now supports parameterized functions.
• Sandboxes: New StreamType controls for Sandbox.exec output handling.
• Sandboxes: The cidr_allowlist argument enables controlled access to specified IP ranges.
• ASGI apps: Lifespan protocol now supported.

🔮 Our founder’s predictions on the AI market

In our latest blog post, Erik shares his vision on where value will accrue in the AI value chain, how patterns in GPU consumption are likely to change, and why Modal is investing in the technology to make GPU supply flexible.

🎨 How to run SD3.5, Flux, Mochi, and more

The never-ending releases are keeping the content team employed!

• Stable Diffusion 3.5: new image diffusion model released just last week by Stability AI, in three variations (Large, Turbo, Medium). Turbo example on Modal.
• FLUX: the other popular set of image diffusion models at the moment. New examples on Modal:
  • Running Flux fast with torch optimizations
  • Running Flux in ComfyUI
  • Fine-tuning Flux with Dreambooth
• Mochi: new text-to-video model. Example on Modal.
• Moshi: new speech-to-speech model. Example on Modal.

We’ve also noticed more users are training models on Modal in addition to deploying them, so we’ve added examples on running long training jobs and parallelizing hyperparameter sweeps.

🍭 Fun tidbits*

• We’re attending AWS re:Invent in December. Let’s meet up!
• We’re hard launching our Youtube channel. Our latest video is on parallelizing a hyperparameter sweep on Modal GPUs.

* or should we say Tidbyt