递归智能发布开源模型 Ripple-65B Preview，在编码基准中表现优异
Ripple-65B Preview delivers strong agentic coding results as a fully Apache 2.0 open US model
Ripple-65B Preview 在 TerminalBench 2.1 编码基准中取得 80.52 分，成为开源美国模型中的最佳成绩。该模型基于 12 万亿 tokens 预训练，使用密集 Transformer 架构，参数量 65B，并针对 Nvidia 硬件优化，部署成本较低，总研发费用仅 3200 万美元。模型在 Hugging Face 上以 Apache 2.0 许可证开源。
同一件事另外 3 家也报道了
- Hacker News · 300+ 热帖··首发
- MarkTechPost··+2 小时 8 分
Ricursive AI 发布 Ripple-65B Preview：一款 65B 密集型开源模型在 Terminal-Bench 2.1 上取得 80.5 分
- IT之家 · AI··+31 小时 58 分
Ricursive Intelligence 发布 Ripple-65B Preview，称其在美国开放权重模型中智能体编程表现最佳
- The Decoder··+2 天·当前阅读
递归智能发布开源模型 Ripple-65B Preview，在编码基准中表现优异
按发布时间排列 · 事件后续出现时自动追加
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.
| Benchmark | Ripple-65B Preview | Inkling 2 Small |
|---|---|---|
| SWE-Bench Pro v2.0 Hard | 79.1% | 86.3% |
| DeepSWE v1.1 | 44.3% | 72.1% |
| MCP Atlas | 71.67% | 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.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
来源: The Decoder · the-decoder.com