[Late catch] Ripple-65B Preview: a 65B dense model tuned for coding agents, with Apache 2.0 weights
What is it?
Ripple-65B Preview is a 65B-parameter dense transformer (no mixture of experts) that Ricursive Intelligence, a company building AI for chip design, trained from scratch. It was pretrained on 12 trillion tokens and accepts up to 1 million tokens of context. The instruction-tuned model (BF16 and NVFP4) and a BF16 base model are on Hugging Face under Apache 2.0.
Why does it matter?
Ricursive reports 80.52Vendor claim on TerminalBench 2.1, ahead of US open-weight models more than ten times its size. That comparison covers US models only, though, and leaves out Chinese models like Kimi and DeepSeek. Because it's a 65B dense model, the company says the NVFP4 checkpoint fits on personal hardware like a DGX Spark. That gives teams one more commercially licensed option for running a coding agent in-house.
Maturity
PreviewIt's a preview. Ricursive says post-training was relatively light, all benchmarks are its own, and no independent results exist yet.
How to track it
- Independent checks: See whether Artificial Analysis or the official Terminal-Bench leaderboard measures Ripple-65B directly.
- Full release: Watch the ricursiveAI page on Hugging Face for a non-preview checkpoint with heavier post-training.