October 2025 Summaries

In a recent update, Modal announced several product enhancements, including the release of Volumes v2 in open beta, providing improved filesystem performance and allowing limitless file storage, ideal for managing large datasets. Additionally, new SDKs for JavaScript/TypeScript and Go are in beta, offering a unified Client object for interacting with Modal resources and extended documentation. Key client updates include App tags for better organization and cost tracking, as well as Sandbox Connect Tokens for secure connections. The announcement also highlighted Nvidia's Flash Attention 4 kernel, which boosts Transformer attention speed, and introduced Modal Vibe, an open-source demo showcasing scalable AI coding platforms. Modal Sandboxes are being utilized by companies like Meta, Cognition, and Restate for various coding and research applications, demonstrating their versatility and scalability. Upcoming events where Modal will be featured include the AI Engineer Code Summit, AWS re:Invent, and NeurIPS 2025, offering opportunities to engage with the team and discuss AI infrastructure advancements.

Modal has partnered with Datalab to enable rapid deployment of high-throughput document intelligence models, notably Marker and Surya, which are designed for efficient document structure parsing and OCR tasks. Marker, a sub-billion-parameter model, offers deterministic and stable document parsing at lower costs compared to larger language models, and has gained significant popularity and trust in the tech community. Modal's infrastructure supports scalable and reliable model serving, allowing developers to deploy these tools quickly, with GPU acceleration for enhanced throughput. These tools are free for research, personal use, and small startups, with commercial licensing available. Marker supports over 90 languages and excels in extracting complex data from PDFs, outperforming other services in accuracy and throughput when deployed on Modal's platform. The partnership emphasizes ease of deployment and scalability, making sophisticated document intelligence accessible to a wider range of users and applications.