humans& introduced Ampersand on October 21 (Beijing time), calling it a collaborator trained to help people achieve their individual and collective goals. Developers can already call it through an API under the model ID ampersand-v0.1.
The company's headline claim is a comparison with Fable: according to humans&, in multi-month simulations, users collaborating with Ampersand accomplished more than with Fable and other frontier models. That result comes from the company's own simulated users, not from an independent benchmark, and humans& did not publish the scores in the announcement post.
The API runs the agent and keeps its conversation history. An agent defines the model, instructions and tools; a session holds one saved conversation that can continue across many turns without resending history. Responses stream as server-sent events, and when the agent wants to use a tool, the app runs the function and returns the result so the same turn can continue. The docs recommend declaring every action, including posting or staying silent in a group chat, as a function tool.
The examples use the official OpenAI Python client (tested with openai 3.24.0) and the OpenAI JavaScript client, pointed at the Ampersand base URL with an API key from the developer console. humans& has not stated per-token prices in the announcement; the docs point to a separate pricing and limits page.