Ricursive releases Ripple-65B Preview with open weights under Apache 2.0

Ricursive Intelligence, a startup that builds AI for chip design, released Ripple-65B Preview on November 4: a 65B dense model it trained from scratch, with the weights public on Hugging Face under Apache 2.0.

WHAT IS OUT

Three repositories under ricursiveAI on Hugging Face, all public and ungated: the instruction-tuned model in BF16 (about 64.8B parameters by the safetensors count), an NVFP4 checkpoint and a DFlash2 drafter. A pretrain-only base model, Ripple-65B Base Preview, has been public there since October 14.

It is also on OpenRouter as ricursiveai/ripple-65b-preview, served by GMI Cloud at $0.20 input and $0.60 output per 1M tokens. The endpoint shows no launch discount, and that listing offers a 16,384-token context window. Ricursive also names Together AI, and says Fireworks is coming.

WHAT RICURSIVE SAYS

Ripple was pretrained on 12 trillion tokens of mostly English text and code, with support for up to 1M tokens of context. On Terminal-Bench 2.1, Ricursive reports 80.52, ahead of Reflection's Beam (501B total, 80.1), Meta's Muse Spark Open (816B, 76.4), Thinking Machines' Inkling (975B, 63.8) and NVIDIA's Nemotron 3 Ultra (550B, 56.4). These are Ricursive's own figures.

Ricursive puts its full research and development cost at $32 million and says the model fits on an NVIDIA DGX Spark. It describes this release as a general-purpose variant of the proprietary model behind its chip design platform, with relatively light post-training focused on agentic coding.

Dense matters for the size comparison: Ripple uses all 65B parameters for every token, while Beam activates 23B of its 501B.

THE SAME DAY

Cognition's SWE-2 Flash also claims to beat every American open-weight model on coding, on Cognition's own evaluations, but its weights are held back for a partner program. Ripple's are downloadable.

WHAT WE DO NOT KNOW

No independent leaderboard has scored Ripple yet. The 1M context Ricursive describes is not what the OpenRouter listing offers today. And there is no date for a release without the Preview tag.