
What happened
Ricursive Intelligence released Ripple-65B Preview under Apache 2.0 on Hugging Face, pretrained on 12 trillion tokens, with a score of 80.52 on TerminalBench 2.1, the best among open US models.
Why it matters
At $32 million covering all research and development, the model may be cheaper to build than comparable efforts, and its dense 65B design is likely easier to deploy on user hardware than Inkling 2 Small.
WHO IT HITSEnterprise IT teams evaluating self-hosted coding models may find Ripple-65B Preview easier to run on their own Nvidia hardware than the larger, higher-scoring Inkling 2 Small.
Summaries like this, every morning.
Ripple-65B Preview stands out not only for its benchmark score but for its reported cost. Ricursive says total development and training costs amounted to just $32 million, covering all research and development expenses.
The model uses a dense Transformer architecture with 65 billion parameters rather than the more common mixture-of-experts approach, and it is optimized for Nvidia hardware. Ripple partly uses Terminus 2 and partly OpenCode as its harness.
While Ripple-65B Preview leads open US models on TerminalBench 2.1, it trails the also newly released Inkling 2 Small on SWE-Bench Pro v2.0 Hard, DeepSWE v1.1, and MCP Atlas. Still, Ricursive argues that Ripple is likely easier to deploy on users' own hardware.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
At a Simeji business meetup held at Google's Shibuya office on October 29, 2026, Simeji operating head Yu Furu…

At the 2026 World Internet Conference Wuzhen Summit, which opened on November 4, leading Chinese tech companie…

In April, Anthropic's Claude Mythos Preview and OpenAI's GPT-5.5 autonomously broke into a simulated 32-step…

A walkthrough of Anthropic's Claude Code shows first launch surfaces a warning that Claude can make mistakes…

The author turned the define-done skill's acceptance criteria into runnable acceptance tests and measured them

On Nov 5, 2026, installing PyPI-stable transformers 5.22.1 failed with 'Transformers does not recognize this a…
