Key points
- Ricursive Intelligence, an AI chip design firm, released Ripple-65B Preview, a 65B-parameter dense Transformer, on Hugging Face under the Apache 2.0 license.
- The company says it scored 80.52 on TerminalBench 2.1, the best result among open US models; it was pretrained on 12 trillion tokens.
- Ricursive says total development and training cost was $32 million, covering all research and development expenses.
- The model trails the newly released Inkling 2 Small on other benchmarks, but may be easier to run on your own hardware.
Why it matters for marketers: Teams building coding agents for marketing automation get another permissively licensed model to test, one that may run on their own hardware instead of relying on closed APIs.
Ricursive Intelligence, a company known for AI chip design, has trained and released an open language model called Ripple-65B Preview. It is available on Hugging Face under the Apache 2.0 license, which allows broad commercial use. According to the company, the model was pretrained on 12 trillion tokens.
Ricursive reports that Ripple scored 80.52 on TerminalBench 2.1, a coding benchmark, which it describes as the best result among open US models. The score relies partly on the Terminus 2 harness and partly on OpenCode. The Decoder notes that Ripple does not match the newly released Inkling 2 Small on other benchmarks, so the result is strong but not across the board.
The architecture is a dense Transformer with 65 billion parameters, not the mixture-of-experts design that is more common, and it is optimized for Nvidia hardware. The Decoder says this likely makes Ripple easier to deploy on users' own hardware. Ricursive also says total development and training costs came to $32 million, covering all research and development.
For marketers and content teams, the relevance is in agentic workflows that depend on coding: scripts that process content, pipelines that connect analytics and publishing tools, and automations built by small technical teams. An open license and self-hosting potential give agencies and founders more control over cost, data handling and customization than a hosted model would.
Because this is a preview and the headline score is company-reported, teams should test it on their own tasks before committing. Watch for independent benchmark runs, and for how Ripple compares with Inkling 2 Small on the workloads that matter to you.
Sources
This brief is our own summary. Read the full reporting at the source.