Ripple-65B Preview
Ricursive Intelligence released Ripple-65B Preview, a 65B dense open-weight model under Apache 2.0 that scores 80.52 on TerminalBench 2.1, which the company says is the top result among U.S. open-weight models.
It is the lab's first public language model, a dense transformer pretrained from scratch on 12 trillion tokens with support for up to 1M context. Ricursive says it trained it for $32M in total R&D cost and reached its score with 1.4M RL rollouts, about 70 times fewer than Reflection Beam. The architecture is thin and deep (d_model 5120, 80 layers) and uses gated attention, global NoPE and value residuals.
- Date
- Wednesday 4 November 2026
- Lab
- Ricursive Intelligence
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| TerminalBench 2.1 | 80.52 Company compares with 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) | company |
| Total R&D cost | $32M Company estimate, covers ladder development through final runs | company |
| RL rollouts | 1.4M About 70x fewer than Reflection Beam, per company | company |
| Pretraining tokens | 12T | company |
Benchmark results are the company's own and have not been independently replicated. License is Apache 2.0 per secondary coverage.
Sources
- ricursive.com/ripple
- pasqualepillitteri.it/en/news/30248/ripple-65b-preview-ricursive-dense-open-weight-llm
- orcarouter.ai/blog/ripple-65b-vs-kimi-k3-1-pro
- mlq.ai/news/ricursive-releases-ripple-65b-preview-an-apache-20-open-model-it-says-cost-32-
This record was checked against its sources on 5 November 2026. How we check