Ricursive Intelligence releases Ripple-65B Preview, a 65B open-weight coding model

Ricursive Intelligence releases Ripple-65B Preview, a 65B dense LLM it says it pretrained from scratch.

The company says the model sets a new state of the art for U.S. open-weight models on agentic coding, ahead of larger models including Muse Spark Open-816B, Beam-501B, Inkling-975B and Nemotron 3 Ultra-550B. These benchmark claims come from the company; no independent results appear in the reports.

The company says the model runs locally on a MacBook Pro or DGX Spark. It is available on Hugging Face, Together AI, GMI Cloud and OpenRouter, served in BF16 with native tool calling.

swyx, quoting the release, says the model cost $32M, one-tenth the cost of the previous open-weight state of the art on agentic coding. A footnote says the $32M covers full R&D, including pretraining ladder development, data mixing, evals, debugging, synthetic data generation, posttraining, and the final runs. Nathan Lambert praises the release as a small-team U.S. open model that follows the OLMo lineage.