New open AI model Ripple-65B beats much bigger rivals at coding

Ricursive Intelligence, an AI lab founded by former Google researchers, has released Ripple-65B Preview. The team says it delivers the best agentic coding performance among US open-weight models.

The lab previously focused on AI-powered chip design tools, and Ripple-65B Preview is the general version of that special model. On the TerminalBench 2.1 test, it beat much larger models like Meta Muse Spark Open 816B and NVIDIA Nemotron 3 Ultra 550B.

Ripple-65B Preview is a dense model trained on 12 trillion tokens. It supports a 1 million token context window, so it can handle very long code and documents.

The company says it cost 32 million dollars to build and train the model from scratch. It shows mid-sized models can still compete at a much lower cost.