Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding
Reported by MarkTechPost, Hacker News, panews.io and 4 more outlets7 sources ↓
Ricursive Intelligence released Ripple-65B Preview, a 65-billion-parameter open-weight model. The company says it scored 80.52 on TerminalBench 2.1, a U.S. open-weight record for agentic coding.
Key points
- Ricursive says the model scored 80.52 on TerminalBench 2.1, above Reflection Beam 501B at 80.1
- The dense model was pretrained on 12 trillion tokens
- Weights are available on Hugging Face under the Apache 2.0 license
- Ricursive says development and training cost $32 million
Sources
- MarkTechPostRead ↗What is Ripple-65B Preview?
- Hacker NewsRead ↗Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding
- panews.ioRead ↗Ricursive Releases Open-Weight Model Ripple-65B Preview, Scoring 80.52 on TerminalBench 2.1
- orcarouter.aiRead ↗Ripple-65B Preview: Apache 2.0 Weights, a 1M-Token Claim and a 16K API
- FourWeekMBARead ↗Ripple-65B Edges a 501B Model 80.5 to 80.1 on Terminal-Bench
Details
Ricursive Intelligence released Ripple-65B Preview as a general-purpose open-weight model. It is a dense Transformer with 65 billion parameters, and Ricursive says it was pretrained on 12 trillion tokens with support for context lengths up to roughly 1 million tokens.
Ricursive reports a score of 80.52 on TerminalBench 2.1, ahead of Reflection Beam 501B at 80.1. The company describes this as the best result among U.S. open-weight models; the reported benchmark result has not been independently confirmed in the articles, and OrcaRouter says Artificial Analysis had no entry for the model as of November 5.
The weights are available on Hugging Face under Apache 2.0, and Ricursive and the Hacker News post list availability through Together AI, GMI Cloud and OpenRouter. The Hacker News post said Fireworks availability was expected later that day. OrcaRouter reports that a commercial endpoint advertised a 16,384-token context limit, despite the model card's claim of up to 1 million tokens.
Ricursive says the full research, development and training cost was $32 million. The Decoder reports that Ripple uses both the Terminus 2 and OpenCode harnesses for its benchmark result, while OrcaRouter notes that the model scored below some newly released models on other benchmarks.
Written from coverage as of
Images
- The Decoder
- orcarouter.ai
- Hacker News
- orcarouter.ai