Ricursive releases Ripple-65B Preview, an open-weight coding model

Ricursive Intelligence released an open-weight model called Ripple-65B Preview on November 4, 2026, according to aitimes.com. The model has 65 billion parameters and is downloadable on Hugging Face, and it is also available through Together AI, GMI Cloud and OpenRouter under the Apache 2.0 license, which permits commercial use, modification and redistribution.

Ricursive reports that Ripple-65B Preview scored 80.52 on the agentic coding benchmark TerminalBench 2.1, ahead of Beam-501B at 80.1, Muse Spark Open-816B at 76.4, Inkling-975B at 63.8 and Nemotron 3 Ultra 550B at 56.4. On SWE-Bench Pro v2.0 Hard it scored 79.1, ahead of Beam at 76.5 and Muse Spark Open at 72.4.

The lead does not hold across every test. On DeepSWE 1.1, Ripple-65B Preview scored 44.3, behind Muse Spark Open at 60.0.

Ricursive estimates the model's total development cost at 32 million USD, about one tenth of the cost of the previous leading U.S. open-weight coding model, Beam. The company says the model was pretrained from scratch on 12 trillion tokens, supports context lengths up to 1 million tokens and uses a dense transformer architecture rather than a mixture-of-experts design. It also says reinforcement learning used only 1.4 million rollouts, about 70 times fewer than Beam.

An NVFP4 version was released for running the model on NVIDIA DGX Spark. In an internal test on two B200 GPUs, Ricursive says the model generated about 250 tokens per second under batched conditions.

Benchmark caveats

VentureBeat noted that benchmark scores can vary with evaluation tools and conditions, so an overall lead among U.S. open-weight models is not established. Ricursive was founded by CEO Anna Goldie and CTO Azalia Mirhoseini, who previously worked on Google's AlphaChip AI semiconductor design technology. It is a separate company from Recursive, which released the coding model Flint.