Overview

Modal is cloud infrastructure for teams that develop, train and serve AI applications. Developers define their code and its environment in Python, and Modal runs it on CPUs or GPUs that spin up in seconds, scaling up and down with demand. The company has announced a Series D funding round and says it has passed $1 billion in annualised revenue run rate.

Products

Modal Inference serves large language models and image, video and audio models, from real-time endpoints to large batch jobs. Sandboxes provide isolated environments for coding agents, reinforcement learning rollouts and untrusted code, and the company has written about scaling to one million concurrent sandboxes. Training lets teams fine-tune models, run reinforcement learning and evaluate models on GPUs without managing clusters. Underneath, Modal says it has rebuilt infrastructure for AI, including its own container runtime, storage and a multi-cloud pool of GPU capacity.

Developer tools

A browser playground lets you try Modal without installing anything. The site also offers a GPU Glossary explaining GPU concepts and an LLM Engine Advisor to help choose a model and serving engine. There is a changelog, a blog, a Slack community and startup credits for eligible companies.

Pricing

Modal charges per second for GPU and CPU usage, so you pay only while code is running rather than for idle servers. Free starting credits are typically offered on sign-up, and enterprise plans are available through sales. Check the pricing page for current GPU rates.

Who it suits

Modal suits machine learning engineers, AI startups and product teams that want to deploy models or run GPU workloads without operating Kubernetes or reserving servers, and agent builders who need safe code execution.

Things to consider

Per-second billing is efficient for bursty workloads, but steady high-volume use can cost more than reserved capacity, so model your expected spend. Cold starts are fast but not zero, which matters for latency-sensitive apps. Review data residency and security options if you handle sensitive data, and set spending limits while you experiment.

Moving code to Modal usually involves adding decorators and environment definitions, so existing Python projects can often be migrated incrementally.