Ricursive Intelligence Releases Ripple-65B Preview Open-Weight Coding Model
Ricursive Intelligence has released Ripple-65B Preview, a 65-billion parameter open-weight dense model trained on 12 trillion tokens. The model features a 1-million-token context length and aims to deliver top-tier agentic programming performance among US open-weight models. Ripple-65B Preview demonstrates competitive agentic programming capabilities on benchmarks like TerminalBench 2.1, outperforming significantly larger models despite its smaller 65B scale. This release highlights the growing competitiveness of smaller, highly optimized open-weight models in specialized software engineering tasks. Developed from scratch at a cost of $32 million, Ripple-65B is a general-purpose variant of the in-house model powering Ricursive Intelligence's AI-driven chip design platform. It is a pretrain-only checkpoint supporting native tool calling and a 1M token context window.
## BACKGROUND
Open-weight models are AI models whose model weights and parameters are made publicly available for research and commercial use, fostering community innovation. TerminalBench 2.1 is an evaluation benchmark used to test large language models on agentic coding and long-horizon tasks.