Recursive unveils 65-billion-parameter AI model… 80.52 points on coding evaluation

Ricursive Intelligence released Ripple-65B Preview, an open-weight AI model with 65 billion parameters that scored 80.52 on Terminal-Bench 2.1, edging out the 501-billion-parameter Reflection Beam 501B's 80.1. The dense model, trained from scratch on 12 trillion mostly English and code tokens across 80 layers, is a general-purpose version of the proprietary model behind Ricursive's AI chip design platform, which the company says can cut chip design cycles to weeks. The result matters because a far smaller open model matching a much larger rival signals rising efficiency in coding-focused AI, and it extends Ricursive's push to tie model development to semiconductor design. What to watch next is whether the preview weights draw outside adoption and whether Ricursive's chip-design claims hold up in production.