Humans&'s group-chat model Ampersand lands on AI Gateway, deciding each turn whether to speak or stay quiet
Synopsis
Ampersand, a language model from Humans&, is now available free on AI Gateway through November 13, 2026. It is built for multi-person conversations such as team chats, deciding on each turn whether to speak or stay quiet, and is designed to understand individual and shared goals and help people work through disagreements. An application exposes speaking and silence as two caller-defined tools, send_message and pass, which the application executes and records; the model accepts auto and none tool choices but rejects required and named ones.
Interpretation
A language model has been placed in group conversations, where its job is not only to produce a reply but to judge on each turn whether speaking now would be useful or whether it should wait for others. Conversational models typically answer every input, whereas Ampersand treats staying quiet as an explicit action for keeping silent while people take turns. The official description's two consecutive human messages show the contrast: the first invites a contribution, the second asks it to wait until Leo replies, and the model chooses between speaking and silence accordingly.
Speaking and silence are split into two caller-defined tools, and the application executes and records the results while the model only picks one. send_message delivers a contribution, while pass keeps the conversation quiet until the next human message; the application executes both actions, and tool choice can be left at its default. The model accepts auto and none but rejects required and named tool choices; calls can go through the AI SDK, the OpenAI-compatible Chat Completions API, or a coding agent, with the model name humansand/ampersand-v0.1.
Perspective
The demonstrated setting is a team-chat-style multi-person conversation: the application exposes send_message and pass as tools, executes and records their results, and the examples cover being invited to contribute and being asked to wait. Deployment requires persisting the response, history and delivery receipts, processing messages sequentially per conversation, and checking conversation access before posting.
The official description reports no evaluation results or comparison with other models, so how accurately the model judges when to speak in real group chats is not established; the free availability is limited to November 13, 2026.