Ricursive releases Ripple-65B Preview, an Apache-2.0 dense open-weight model aimed at agentic coding
The 65B dense model (12T pretraining tokens, up to 1M context) self-reports 80.5 on Terminal-Bench 2.1, ahead of much larger US open-weight MoEs, and ships BF16 plus an experimental NVFP4 checkpoint sized for a DGX Spark (Hugging Face). It's also served on Together AI at $0.20/$0.60 per million input/output tokens.