Vercel Drives, persistent storage mounted as a directory inside a Vercel Sandbox, is now generally available on all plans. Files survive after the sandbox stops, so agents can keep a workspace or file-based memory across runs, and drives can be forked for parallel workers.
Apple researchers propose Selection-based Structured Reasoning (SSR), in which a multimodal agent picks from a set of predefined reasoning steps instead of writing free-form reasoning before each action. The aim is to cut the inference cost of long, unhelpful reasoning in small models.
Apple researchers compare two ways for LLM agents to find the right tools in large registries: retrieving tools similar to the user query, or first having the model sketch the function signatures it needs and searching with those. The question matters because context windows cannot hold every tool definition.
Together AI now hosts Meta's open-weight Muse Spark Open, a model aimed at agentic coding, with a 1-million-token context window. It is offered both serverless and on dedicated infrastructure.
Vercel's AI Gateway now serves Meta's open-weight Muse Spark Open, a reasoning model for coding and agent workflows that accepts text and images with a 1-million-token context. It can be used from coding agents such as Claude Code, Codex and Cursor, with zero data retention.
An unattributed reasoning model called Cat Jam is available for free on Vercel's AI Gateway during a stealth period. It targets coding and agent tasks with function calling and structured outputs, takes text only, and prompts may be retained for safety reasons.