Ripple-65B Preview is a 65B dense LLM pretrained from scratch by Ricursive Intelligence. It sets a new state-of-the-art for U.S. open-weight models on agentic coding tasks, outperforming much larger models, with scores of 80.5% on TerminalBench 2.1, 79.1% on SWE-Bench Pro v2.0-Hard and 44.3% on DeepSWE v1.1. Trained on 12 trillion tokens with 1M context support, the model fits on a MacBook or DGX Spark. Weights and inference are available now via Hugging Face, GMI Cloud and OpenRouter; Together AI and Fireworks access follows later the same day. A full technical report is forthcoming.
Ricursive Intelligence released Ripple-65B Preview as a general-purpose variant of the language model powering its AI chip-design platform. The model is a 65-billion-parameter dense Transformer pretrained from scratch on 12 trillion tokens, with support for up to 1 million tokens of context. Ricursive reports a TerminalBench 2.1 score of 80.52 and describes the model as optimized for agentic coding and efficient training and serving on NVIDIA hardware. The release includes instruction-tuned BF16 and NVFP4 checkpoints, as well as a BF16 base model.
