Ripple-65B Preview is an open-weight language model from Ricursive Intelligence, aimed at agentic coding (AI that plans, writes and runs code over many steps) and small enough for consumer hardware such as NVIDIA's DGX Spark.
In plain words
Ricursive Intelligence, an AI lab working on chip design, released Ripple as a general-purpose version of its chip design model. It was trained from scratch on 12 trillion tokens of mostly English text and code. It is a dense Transformer with 65 billion parameters, not a mixture-of-experts model, and reads up to 1 million tokens of context. Ricursive reports 80.52 on TerminalBench 2.1, the best US open-weight score, for a full research cost of $32M. The weights are on Hugging Face under Apache 2.0.
A quick example
Imagine a developer who wants a coding agent that never sends code to an outside service. They download an open-weight model of this size, run it on a desktop machine, and point their agent tool at it to fix a failing test.
Where you'll meet it
- Coding agents that need an open-weight model.
- Running a model on local hardware such as DGX Spark.
Related
Sources
- DocsIntroducing Ripple-65B Preview, Ricursive Intelligence · read 5 Nov 2026
- DocsricursiveAI/Ripple-65B-Preview model card, Hugging Face (Ricursive Intelligence) · read 4 Nov 2026
Checked by WhatIsAI against the sources above · 5 Nov 2026. Spotted an error? Report it.