Ricursive Intelligence opens the weights of Ripple-65B Preview, a 65-billion-parameter model trained from scratch
Chip design AI startup Ricursive Intelligence has released Ripple-65B Preview, with weights on Hugging Face under Apache 2.0, in an instruction-tuned version and a pretrain-only base version. The 65-billion-parameter model uses all of its parameters at every step. It was trained from scratch on 12 trillion tokens of text and code and handles up to 1 million tokens of context. The company says it is a general-purpose version of the model behind its chip design platform, that the team grew from 1 to 5 people during development, and that building it from scratch cost about $32 million in total. The company reports that it scores 80.52 on the agentic coding test TerminalBench 2.1. It runs on desktop devices such as NVIDIA DGX Spark and is also on Together AI, GMI Cloud and OpenRouter.
Why it matters
A team of at most five people trained a 65-billion-parameter open-weight model from scratch for about $32 million, giving groups without big-lab budgets a concrete reference for cost and method.
Sources1
Written by dalao.ai: the title, summary and why it matters are our own words. The original reports are the record; their publishers keep authorship.