Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding
Summary
Ricursive Intelligence unveils Ripple-65B Preview, an open-weight LLM optimized for agentic coding with 1M context and claimed SOTA performance. The article details a dense Transformer architecture, cost-efficient pretraining at $32M, and a laddered RL/SFT approach with synthetic data, aiming for efficient on-device inference and strong code capabilities, while highlighting open-source contributions and multi-platform availability.