Ricursive says it trained an open-weight coding model for $32m
Ripple-65B Preview scored 80.52 on TerminalBench 2.1, ahead of Reflection Beam 501B at 80.1, on the company’s figures. Moonshot and Tencent also released open weights.
Source published · In the Daily Brief of 5 November 2026
Filed under Large language models, AI industry
Ricursive Intelligence, which builds AI for chip design, said Ripple-65B Preview was pretrained on 12 trillion tokens and supports a context of up to 1m tokens. The company puts its full research and development cost at $32m, which it estimates at one-tenth of the previous open-weight leader on agentic coding. It says it reached its score with 1.4m rollouts, roughly 70 times fewer than Reflection Beam.
The benchmark scores and cost comparison are Ricursive’s own. Moonshot AI published the weights of Kimi K3.1 Pro, post-trained from Kimi K3 with a context window of 1m tokens. Tencent released Hy4, which it says joins the top tier of Chinese open-weight models, according to Bloomberg.