Mouse-1

Mouse-1 is a small decision model that sits inside your agent harness. An agent makes hundreds of small choices during a task, and Mouse-1 takes the agent's state and options and returns a probability for each, so the harness can settle the small calls cheaply while the frontier model handles the reasoning and code. The team says it is 163x cheaper and 65x faster than a frontier model for these calls. Available as an API.