Physix Frontier · News Briefing Card (IT Home · Nov 6, 2026)

Ricursive releases Ripple-65B Preview open-weight model

KEY FACTS

Ricursive Intelligence released the Ripple-65B Preview model on November 4.
The model uses a dense architecture, was pretrained on 12T tokens, and supports a 1M context length.
The company says it delivers the best agentic coding performance among US open-weight models.
On TerminalBench 2.1, it outperforms larger models from Meta and NVIDIA.
Developing and training the model from scratch cost $32 million.

KEY DATA

12TPretraining tokens

1MContext length

$32 millionDevelopment and training cost

PHYSIX OBSERVATION

A small team spent $32 million to build a coding-agent model that rivals Big Tech's ultra-large models, showing that competition in the open-weight track is shifting from piling on parameters to competing on data and training efficiency. This is good news for developers, but "best in the US" is only a self-claim and should not be taken at face value until third-party retesting.

Source: IT Home report