Breaking

Ricursive AI's Ripple-65B Preview Now Available on Design Arena, Claimed Top Performance Among US Open-Weight Models

November 4, 2026 at 16:51 EST

Foundation Models
Software Dev & Coding

Ricursive AI's Ripple-65B Preview is now available on Design Arena, with the company positioning the general-purpose model for agentic coding and complex tasks and claiming state-of-the-art performance among US open-weight models.

Ricursive AI · Ripple-65B Preview

A general-purpose open-weight model aimed at agentic coding, with a 1-million-token context window

Ricursive AI is positioning Ripple-65B Preview for agentic coding and complex tasks, and claims state-of-the-art performance among US open-weight models. Those performance results have not yet been independently confirmed.

12T

tokens used in pretraining

token context window in a single session

#1?

claimed rank among US open-weight models (unverified)

Pretraining scale vs. context window

Each block = 1 trillion tokens of pretraining data. The context window measures how much the model can read at once, a different quantity, so it is shown to scale.

12T

pretraining tokens

context window (about 1/12,000,000 of the block height, shown as a sliver)

Takeaway: the two numbers measure different things. The 12T figure describes training data; the 1M figure describes how much code or text a developer could load into one session.

How agentic coding works

The model is aimed at multi-step work with limited human intervention, not only single code suggestions.

Step 1

Plan

Break a task into multi-step work

→

Step 2

Use tools

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.