Ex-Googlers' Ricursive Intelligence Unveils Ripple-65B Preview: A New Open-Weight SOTA for Agentic Coding

Ricursive Intelligence, a startup founded by former Google Brain engineers, unveiled Ripple-65B Preview on November 4, 2026, a 65-billion-parameter dense open-weight model that claims U.S. open-weight state-of-the-art for agentic coding, achieving a score of 80.52 on TerminalBench 2.1. For broader context, explore our AI Tools Pricing.

Introducing Ripple-65B Preview

The Ripple-65B Preview is a dense Transformer model that Ricursive Intelligence trained from scratch using 12 trillion tokens. It supports an extensive context length of up to 1 million tokens, a feature designed to enhance its capabilities in complex coding tasks. The model's architecture and training methodology aim to optimize performance for agentic workflows.

Benchmark Performance and SOTA Claims

Ricursive Intelligence reports that Ripple-65B Preview achieved a score of 80.52 on TerminalBench 2.1, a benchmark used to evaluate agentic coding performance. This score underpins the company's claim of establishing a new U.S. open-weight SOTA in this domain. Additionally, the model reportedly leads on SWE-bench Pro v2.0 Hard, further solidifying its position in advanced coding tasks.

Cost-Efficiency in Development

A notable aspect of Ripple-65B's development is its reported cost-efficiency. The total development cost for the model was $32 million, which Ricursive Intelligence states is approximately one-tenth the cost of the previous open-weight agentic-coding SOTA. This cost reduction is attributed, in part, to its reinforcement learning (RL) training, which required about 70 times fewer rollouts than the Reflection Beam model.

Comparative Development Costs

• Ripple-65B Preview: $32 million
• Previous Open-Weight Agentic-Coding SOTA: Approximately $320 million (based on Ricursive Intelligence's claim of one-tenth the cost)

This cost difference highlights a potential shift in the economic feasibility of developing high-performance open-weight models.

Availability and Licensing

Ripple-65B Preview is available under the Apache 2.0 license, making it accessible for broad use and further development. Users can access the model through various platforms, including Hugging Face, Together AI, GMI Cloud, and OpenRouter. This wide distribution aims to foster adoption and integration into diverse AI applications.

Ricursive Intelligence's Focus

Ricursive Intelligence operates as a full-stack lab, concentrating its efforts on co-optimizing hardware, software, and models specifically for AI-driven chip design. The release of Ripple-65B Preview aligns with this mission, showcasing the company's capabilities in developing advanced AI models that could influence future hardware and software innovations.

Conclusion

The introduction of Ripple-65B Preview by Ricursive Intelligence marks a significant development in the field of open-weight agentic coding models. Its reported performance on key benchmarks, coupled with a substantially lower development cost, positions it as a notable advancement. The model's availability across multiple platforms under an open-source license suggests a commitment to broader accessibility and collaborative innovation in AI-driven development.

Sources

• ricursiveAI/Ripple-65B-Preview · Hugging Face
• ricursiveAI/Ripple-65B-Preview-NVFP4 · Hugging Face
• AtomicChat/Ripple-65B-Preview-GGUF · Hugging Face
• model-00010-of-00016.safetensors · ricursiveAI/Ripple-65B-Preview at main
• How Ricursive Intelligence raised $335M at a $4B valuation in 4 months | TechCrunch