Published

Chip-design firm Ricursive releases Ripple-65B Preview: Apache 2.0, 65B dense, 80.52 on TerminalBench 2.1

AI chip-design company Ricursive Intelligence has released Ripple-65B Preview, an open-weight model under the Apache 2.0 license. The 65B-parameter dense model supports up to 1M tokens of context, and the company says its 80.52 on TerminalBench 2.1 is the top score among US open-weight models; The Decoder notes it trails Inkling 2 Small on other coding and tool benchmarks.

White RICURSIVE INTELLIGENCE wordmark on black with the line Recursive Self-Improvement via AI for Chip Design & Chip Design for AI
Share image from Ricursive Intelligence’s launch page.

According to Ricursive’s model card, Ripple-65B Preview is a 65-billion-parameter dense Transformer, not a mixture-of-experts model, pretrained from scratch on 12 trillion tokens with up to 1,048,576 tokens of context. The release includes instruction-tuned models in BF16 and NVFP4 plus a BF16 base model, under Apache 2.0. The company describes it as a general-purpose variant of the proprietary model behind its chip-design platform.

Ricursive says it scores 80.52 on TerminalBench 2.1, ahead of Reflection Beam 501B (80.1), Meta Muse Spark Open 816B (76.4), Thinking Machines Inkling 975B (63.8) and NVIDIA Nemotron 3 Ultra 550B (56.4), and that its RL stage used only 1.4 million rollouts, about 70 times fewer than Reflection Beam. Other results listed by The Decoder are weaker: 79.1% on SWE-Bench Pro v2.0 Hard, 44.3% on DeepSWE v1.1 and 71.67% on MCP Atlas, all below the also newly released Inkling 2 Small.

The company says the model runs on local hardware such as NVIDIA DGX Spark and generates about 250 tokens per second on two B200s at batch size 8; according to The Decoder, Ricursive puts total R&D and training costs at about $32 million. The model is on Hugging Face and available through Together AI, GMI Cloud, OpenRouter and Fireworks. As a Preview it has had relatively light post-training focused on agentic coding.

Sources

  1. Ripple-65B Preview, a Frontier Open-Weight Model | Ricursive Intelligence — Ricursive Intelligence
  2. ricursiveAI/Ripple-65B-Preview · Hugging Face — Ricursive Intelligence (Hugging Face)
  3. Ripple-65B Preview delivers strong agentic coding results as a fully Apache 2.0 open US model — The Decoder
Open this story in Shift →