Ricursive releases open-weight model Ripple-65B Preview

Ricursive Intelligence released Ripple-65B Preview, an open-weight model with a 650 billion parameter dense architecture that was pretrained from scratch and supports contexts of up to 1 million tokens. The company said the model scored 80.52 on the agentic coding benchmark TerminalBench 2.1, edging past Reflection Beam 501B's 80.1. Ricursive put total R&D and training costs at $32 million and published the weights on platforms including Hugging Face. The release matters because it delivers frontier-class agentic coding performance in an open-weight package at a relatively modest training budget, intensifying competition among open model developers. What to watch next is how the open-weight community adopts and fine-tunes Ripple-65B, and whether its benchmark lead holds up in independent evaluations.