Ricursive releases open-weight model Ripple-65B Preview

On November 5, according to an official announcement from Ricursive Intelligence, the company released the open-weight model Ripple-65B Preview. The model features a 65-billion parameter dense architecture, is trained from scratch, and supports a context window of up to 1 million tokens.

Ricursive stated that the model scored 80.52 on the agentic programming benchmark TerminalBench 2.1, surpassing Reflection Beam 501B's score of 80.1. The company noted that the total R&D cost for development and training was $32 million, and the weights are now available on platforms such as Hugging Face.