Humans& ships Ampersand, a model built to stay silent in group chats
Ampersand v0.1 reads every message in a group conversation and decides, turn by turn, whether to reply or stay quiet. It is free on Vercel AI Gateway through 13 November.
Humans&, the AI lab founded by former OpenAI, xAI and Anthropic researchers, has shipped its first model. Vercel’s AI Gateway added Ampersand v0.1 on 6 November, free to call through 13 November under the id humansand/ampersand-v0.1.
The model is built for a problem most chat assistants ignore: it reads every message in a conversation, including ones not addressed to it, and after each one decides whether to speak or stay quiet, rather than replying to everything. It does this through two tool calls rather than a system prompt instruction.
// on each turn, the model picks one
send_message({ text: '...' })
pass()Ampersand is built on Z.ai’s open-weight GLM 5.3, trained with reinforcement learning across simulated multi-person dialogues running up to months of simulated time, and carries a 1 million token context window. It’s reachable through the AI SDK, an OpenAI-compatible Chat Completions API, and coding agents including Claude Code, Codex and Cursor.
Why it matters: a bot that answers every message in a team channel gets muted fast. A model trained specifically to judge when to stay out of a conversation is a real building block for anyone putting an agent into Slack or Teams, not just a chat API with instructions bolted on.
The caveat: this is a v0.1 from a lab that hasn’t shipped anything else publicly, and the free access on Vercel ends in a week.