Ricursive Intelligence Releases Ripple-65B Preview Open-Weight Model
Ricursive Intelligence has released Ripple-65B Preview, an open-weight model pretrained on 12 trillion tokens with a 1M context window.
1 source 23m ago
Reporting from 1 source: GIGAZINE.
Why it matters
Ripple-65B Preview shows a US startup buying benchmark parity with far larger open-weight models through training efficiency rather than scale, with a stated development bill of $32 million.
Key Facts
- Ricursive Intelligence released the open-weight model Ripple-65B Preview.
- Ripple-65B Preview scored 80.52 on the agentic coding benchmark Terminal-Bench 2.1.
- Beam 501B scored 80.1, Muse Spark Open scored 76.4 and Inkling 975B scored 63.8 on the same benchmark.
- Ripple-65B Preview has 65 billion parameters and was pretrained on 12 trillion tokens.
- Ricursive Intelligence says developing Ripple-65B Preview from scratch cost $32 million.
It scored 80.52 on Terminal-Bench 2.1, above Beam 501B (80.1), Muse Spark Open (76.4) and Inkling 975B (63.8), despite having 65 billion parameters. Ricursive says developing the model from scratch cost $32 million, about one-tenth the cost of the model that previously held the best open-weight score.
Ricursive Intelligence says it spent $32 million developing Ripple-65B Preview from scratch, including data curation, evaluation, synthetic data generation and the final training run. The company attributes the low figure to computational efficiency and an in-house synthetic data pipeline rather than purchased post-training data.
The 65-billion-parameter model is designed for efficient training and serving on NVIDIA hardware, and its small parameter count lets it run on hardware beyond data-center GPUs. On the math, safety and automation benchmarks shown by Ricursive, it sits third behind Muse's Spark Open and Beam 501B.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google