Ricursive Releases Ripple-65B Preview, an Apache 2.0 Model It Says Tops US Open-Weight Coding Scores for $32 Million
Ricursive Intelligence, a lab building AI for chip design, released Ripple-65B Preview on November 4, a 65-billion-parameter dense model under the Apache 2.0 licence. It reports 80.52 on TerminalBench 2.1, ahead of several far larger US open-weight models, and says the whole project cost $32 million. All benchmark figures are the company's own.
A 65-billion-parameter model from a chip-design startup is claiming coding scores that US labs needed models eight to fifteen times larger to reach. Ricursive Intelligence released Ripple-65B Preview on November 4 under the Apache 2.0 licence, saying it sets a new state of the art among US open-weight models for agentic coding. The claim is the company's own and has not been independently checked.
What Ricursive released#
Ripple-65B Preview is a dense model, not a mixture-of-experts, pretrained from scratch on 12 trillion tokens with up to a 1 million-token context window. Ricursive says it is a general-purpose variant of the proprietary model behind its AI chip design platform, and that it is small enough to run on local hardware such as Nvidia's DGX Spark. It is on Hugging Face, Together AI, GMI Cloud and OpenRouter.
On TerminalBench 2.1, Ricursive reports 80.52, against figures it cites of 80.1 for Reflection Beam (501B), 76.4 for Meta's Muse Spark Open (816B), 63.8 for Thinking Machines' Inkling (975B) and 56.4 for Nvidia's Nemotron 3 Ultra (550B). On X it also reported 79.1% on SWE-Bench Pro v2.0-Hard and 44.3% on DeepSWE v1.1.
The cost claim#
Ricursive says it built and trained the model for $32 million, counting its full research and development from scratch, including data work, evaluations, synthetic data and the final training runs. It calls that one-tenth the cost of the previous open-weight leader on agentic coding, "as estimated to the best of our knowledge," and says it reached its score with about 1.4 million reinforcement-learning rollouts. Reflection, whose Beam model Ricursive compares against, unveiled Beam in October as a 501-billion-parameter sparse model.
Limits#
- Every benchmark number and the cost figure are self-reported; we did not check them on any public leaderboard.
- The comparison scores for other models are as cited by Ricursive and may use different harnesses or settings.
- The cost comparison with other labs rests on Ricursive's estimates of their spending.