Illustration, AI-generated from the story text. Tap a panel to read the story through it.
Ricursive Intelligence released Ripple-65B Preview under , a dense model that scored 80.52 on 2.1 by its own count, against 80.1 for Reflection's 501-billion-parameter Beam and 76.4 for Meta's Muse Spark Open. It puts the full research and development cost at $32 million.
Its own warns that replicating scores "has been challenging" and that results were generally lower on the official , so the lead is a claim until someone else runs it.
If it holds, 65 billion landed 0.4 points from a model nearly eight times the size.
Ricursive's 65-billion-parameter Ripple claims the US open-weight lead in agentic coding at $32 million
Ricursive Intelligence released Ripple-65B Preview under Apache 2.0, a dense model that scored 80.52 on Terminal-Bench 2.1 by its own count, against 80.1 for Reflection's 501-billion-parameter Beam and 76.4 for Meta's Muse Spark Open. It puts the full research and development cost at $32 million. Its own model card warns that replicating scores "has been challenging" and that results were generally lower on the official harness, so the lead is a claim until someone else runs it. If it holds, 65 billion parameters landed 0.4 points from a model nearly eight times the size. SourcesAA
Sources
- ARipple-65B Preview (Ricursive Intelligence)ricursive.com
- AricursiveAI/Ripple-65B-Preview (Hugging Face)huggingface.co
This story ran in the Afternoon Brief of November 4th 2026.