Last updated: November 6, 2026.
Ricursive Intelligence released Ripple-65B Preview, a 65-billion-parameter open-weight model focused on reasoning and agentic coding with support for context windows up to one million tokens.
Quick answer: Ripple-65B Preview is a dense 65B open-weight model trained from scratch on 12 trillion tokens. Ricursive reports a TerminalBench 2.1 score of 80.52 and launch availability through Hugging Face, Together AI, GMI Cloud and OpenRouter.
What is Ripple-65B Preview?
Ripple-65B Preview is a general-purpose variant of a proprietary model used inside Ricursive Intelligence’s AI chip-design platform. Unlike many recent frontier systems that use Mixture-of-Experts architectures, Ripple-65B is a dense Transformer.
Ripple-65B specs
| Specification | Ripple-65B Preview |
|---|---|
| Parameters | 65B |
| Architecture | Dense Transformer |
| Training tokens | 12 trillion |
| Context | Up to 1M tokens |
| Layers | 80 |
| Weights | Open-weight preview |
How good is Ripple-65B at coding?
Ricursive reports a score of 80.52 on TerminalBench 2.1, a benchmark focused on agentic coding. The company says this was a new U.S. open-weight state of the art at launch.
Why the 65B size matters
At 65B parameters, Ripple is smaller than many open-weight frontier systems with hundreds of billions of parameters. Ricursive specifically highlights efficient serving and local-device use cases such as NVIDIA DGX Spark.
How much did it cost to train?
Ricursive says the complete research and development cost was approximately $32 million. The company presents the result as evidence that careful architecture and training can reduce the cost of reaching competitive coding performance.
Where can you try Ripple-65B?
Ricursive lists Hugging Face, Together AI, GMI Cloud and OpenRouter among launch providers, with additional provider availability rolling out.
Why 1M context matters
Support for up to one million tokens is useful for large software repositories, long-running agents and technical workflows that need to keep extensive context available in one session.
FAQ
Is Ripple-65B open source?
Ricursive describes it as open-weight. Open-weight does not necessarily mean the full training stack and dataset are open source.
How many parameters does Ripple have?
65 billion parameters.
What is the context window?
Up to one million tokens.
Sources
Explore more: AI Models – Latest AI Tools
