[Ricursive Releases Open-Source Weight Model Ripple-65B Preview, Achieves 80.52 on TerminalBench 2.1] According to an announcement by Ricursive Intelligence, the company has released Ripple-65B Preview, a 65B dense open-source weight model pre-trained from scratch. It supports contexts with up to approximately 1 million tokens and achieved a score of 80.52 on the agent programming benchmark TerminalBench 2.1, surpassing larger-scale US open-source weight models such as Reflection Beam 501B (80.1). The total development cost was approximately $32 million, and the weights are already deployed on Hugging Face, Together AI, GMI Cloud, and OpenRouter.