Ricursive AI releases Ripple-65B Preview, a 65B dense open-weight model for agentic coding

Ricursive AI released Ripple-65B Preview on November 4, a 65-billion-parameter dense language model it says it pretrained from scratch. The company claims it sets a new state of the art for U.S. open-weight models on agentic coding tasks, ahead of much larger models it names: Muse Spark Open-816B, Beam-501B, Inkling-975B and Nemotron 3 Ultra-550B. These are the company's own claims.

Ricursive says the model was pretrained on 12 trillion tokens, supports up to 1M tokens of context, and scores 80.52 on Terminal-Bench 2.1 against 80.1 for Reflection's Beam 501B. It puts its full research and development cost at $32 million. The company says the model is on Hugging Face, Together AI, GMI Cloud and OpenRouter.

Shayan Talaei, one of the model's authors, says on Terminal-Bench 2.1 it is a new U.S. open-weight best. He says a core team of five researchers built it, and that they chose a dense design for low latency over long sequences of model and tool calls and a simpler training and serving stack. He says they bought no post-training data, building their own data generation and filtering pipelines on open-source training and inference stacks, and tuned the reinforcement-learning system and task curriculum together.