Modal

Modal is a cloud platform for building and running data, AI, and machine learning applications. It provides serverless compute for GPU inference, sandboxes, training, and batch workloads, with usage-based pricing where you only pay for the time your code is running. The company built its own infrastructure, including a custom file system, container runtime, scheduler, and container image.

Modal's product areas include Inference, Sandboxes, Training, Notebooks, and Batch. Inference supports serving LLMs, image, video, and audio models from real-time to batch, with shared endpoints, dedicated endpoints, custom model deployment, and support for serving stacks such as SGLang and vLLM. Sandboxes provide isolated environments for coding agents, reinforcement learning rollouts, and untrusted code. Training covers fine-tuning, reinforcement learning, and model evaluation on GPUs. The platform also includes a Playground for browser-based exploration, an LLM Engine Advisor, and a GPU Glossary.

Modal is aimed at teams building generative AI inference, LLM fine-tuning, computational biotech, and media processing applications. Pricing is usage-based, with plans including a Starter plan for small teams and independent developers, a Team plan for startups and larger organizations, and an Enterprise plan with custom terms. Plans include free compute credits, with additional credit grants available for startups and academics.