January 2026 Summaries

Modal announced availability through AWS and Google Cloud Marketplaces for enterprise customers, allowing purchases and usage management through existing cloud spend commitments, alongside sandbox observability improvements that add clearer resource, region, and execution-lifecycle details. Recent client releases introduce Python 3.14 and experimental free-threaded Python support, enhanced CLI token and timestamp inspection, and warnings for potentially blocking API calls in asynchronous contexts. The update also highlights Ramp’s use of Modal sandboxes to run its internal Inspect coding agent, which produces roughly 30% of the company’s production pull requests, and shares guidance on optimizing offline, online, and semi-online LLM inference workloads with engines such as vLLM and SGLang. Modal additionally describes its practices for maintaining a distributed fleet of more than 20,000 GPUs, including instance testing, image preparation, health checks, and reliability monitoring, while promoting February events in New York and San Francisco focused on AI, startups, reinforcement learning, and biotech.

Chai Discovery, a machine-learning drug discovery company, uses Modal to support flexible, large-scale computational workflows for designing medicines across biological targets and modalities. Its pipelines combine heterogeneous models, large biological datasets, and GPU-intensive tasks that can rapidly expand from small experiments to thousands of inference jobs, making traditional cloud infrastructure costly and operationally complex. Modal provides reproducible execution environments, shared persistent storage through Modal Volumes for large multiple sequence alignment datasets, and elastic GPU scaling that allows Chai to launch hundreds of GPUs within minutes and scale down when demand subsides. By avoiding repeated data downloads, hardware inconsistencies, manual cluster management, and infrastructure rewrites, Chai can use the same Python-based workflows from early research through production, accelerating experimentation and allowing scientists to focus more directly on molecular discovery.