Many MindedRipple-65B Preview: Open-Source Coding Model Shows Promise But Needs Refinement
Ricursive Intelligence published Ripple-65B Preview on November 6, 2026, as a fully open-source US language model under Apache 2.0 licensing. The model, pretrained on 12 trillion tokens, achieved 80.52 on TerminalBench 2.1 and 79.1% on SWE-Bench Pro v2.0 Hard—indicating solid coding capabilities. It also scored 71.67% on MCP Atlas and 44.3% on DeepSWE v1.1, though it underperforms compared to Inkling 2 Small. With 65 billion parameters and optimized for Nvidia hardware, Ripple-65B Preview is designed to deploy more easily on user hardware than competitors. Total development costs reached $32 million.
The model’s open license and hardware efficiency could lower barriers for developers building tools that address knowledge gaps in technical fields. If scaled to production, it might reduce the cost of coding assistance for individuals and small teams, potentially freeing up resources for other learning needs. However, its current status as a 'Preview' model means it lacks real-world stability and hasn’t yet matched industry benchmarks.
What matters next: Whether Ripple-65B Preview closes the performance gap with Inkling 2 Small through iterative updates. The model’s $32 million cost and 12 trillion-token training scale suggest significant resources were invested—yet its current limitations mean it won’t deliver immediate abundance. For knowledge access to become truly free, this model must transition from a promising preview to a reliable production tool without compromising its open licensing.
Source: The Decoder