Ricursive 发布 Ripple-65B Preview 开源模型 智能体编程表现亮眼

作者:Matthias Bastian原标题:Ripple-65B Preview delivers strong agentic coding results as a fully Apache 2.0 open US model
74AI 研判分

AI 芯片设计公司 Ricursive Intelligence 训练并开源了 Ripple-65B Preview,该模型基于 12 万亿 token 预训练,采用 65B 稠密 Transformer 架构而非混合专家,在 TerminalBench 2.1 编程基准上取得 80.52 分,为美国开源模型最佳成绩。模型以 Apache 2.0 许可发布在 Hugging Face,官方称全部研发训练成本仅 3200 万美元,但综合基准仍不及同期发布的 Inkling 2 Small。

报道正文

Ripple-65B Preview delivers strong agentic coding results as a fully Apache 2.0 open US model

文本
  [Matthias Bastian](https://the-decoder.com/author/matthias-bastian/) [View the LinkedIn Profile of Matthias Bastian](https://www.linkedin.com/in/matthias-bastian-128b71b1/)   Nov 6, 2026          

AI chip design firm Ricursive Intelligence has trained an open language model. Ripple-65B Preview was pretrained on 12 trillion tokens and, according to the company, achieved a score of 80.52 on the TerminalBench 2.1 coding benchmark, the best result among open US models. Ripple partly uses Terminus 2 and partly OpenCode as its harness. Other benchmark results show that Ripple does not match the also newly released Inkling 2 Small, but Ripple is likely easier to deploy on users' own hardware. According to Ricursive, total development and training costs amounted to just $32 million, covering all research and development expenses.

BenchmarkRipple-65B PreviewInkling 2 Small
SWE-Bench Pro v2.0 Hard79.1%86.3%
DeepSWE v1.144.3%72.1%
MCP Atlas71.67%85.4/85.0%

The model uses a dense Transformer architecture with 65 billion parameters instead of the more common mixture-of-experts approach and is optimized for Nvidia hardware. Ripple-65B Preview is available on Hugging Face under the Apache 2.0 license.

AdAd