MCP App StoreLe meilleur de l'écosystème IA et MCP, sélectionné chaque jour.
Yesterday focused on the intersection of efficiency and scale, from high-performance training kernels to compact models designed for edge deployment. The most notable release was Liquid AI s LFM2.5-2.6B, which pushes high-reasoning capabilities into a small enough footprint for local agent deployment on edge devices.
On the infrastructure side, Cursor contributed to the open-source ecosystem by releasing Mixture-of-Kittens, a megakernel designed to optimize MoE training on NVL72s by fusing communication and computation. Meanwhile, Anthropic shifted focus toward the enterprise side, releasing detailed guidance on cost visibility and control for Claude to help admins manage API spending and optimize prompts.
Today s stories:
Overall, the day highlights a continuing trend toward optimizing the cost and resource footprint of AI, whether at the training scale or the local edge.
Liquid AI releases LFM2.5-2.6B, a compact model optimized for deploying local agents. It aims to bring high-reasoning capabilities to edge devices with minimal resource overhead.
Cursor open-sources Mixture-of-Kittens, a deterministic MoE training megakernel for NVL72s. It fuses communication and computation into a single kernel to optimize large-scale model training.
Anthropic provides a comprehensive guide for IT admins to manage Claude spending. Key features include spend caps, usage analytics, and API optimizations like prompt caching.