What changed
Ricursive Intelligence announced the Ripple-65B Preview, an open-source language model trained on 12 trillion tokens. The model achieved a score of 80.52 on the TerminalBench 2.1 coding benchmark, the best result among open US models. It uses a dense Transformer architecture with 65 billion parameters and is optimized for Nvidia hardware.
What this means in practice
The Ripple-65B Preview is available on Hugging Face under the Apache 2.0 license, making it accessible for developers and researchers. According to Ricursive, the model is likely easier to deploy on users' own hardware compared to other recently released models like Inkling 2 Small. The model uses Terminus 2 and OpenCode as its harness, and its dense Transformer architecture differs from the more common mixture-of-experts approach.
Limitations and availability
Benchmark results show that Ripple-65B Preview does not match the performance of Inkling 2 Small on certain tests, such as SWE-Bench Pro v2.0 and DeepSWE v1.1. The model's total development and training costs amounted to $32 million, covering all research and development expenses. The Ripple-65B Preview is available on Hugging Face, but the supplied announcement does not specify details about deployment requirements beyond its optimization for Nvidia hardware.
