A 65B dense model just claimed the U.S. open coding lead: Ripple-65B Preview 🌊

Most of the open coding race this year has been about bigger Mixture-of-Experts stacks. More total parameters, more experts, more serving complexity. This week a small U.S. lab took the opposite bet, and I like it a lot. Ricursive Intelligence released Ripple-65B Preview: a dense 65-billion-parameter model, pretrained from scratch on 12 trillion tokens, with weights on Hugging Face under Apache 2.0.

What they claim

On their model card and launch page, Ripple scores 80.52 on Terminal-Bench 2.1, just ahead of Reflection Beam 501B at 80.1, and well ahead of Meta Muse Spark Open 816B (76.4), Thinking Machines Inkling 975B (63.8) and NVIDIA Nemotron 3 Ultra 550B (56.4) in that comparison. On SWE-bench Pro v2.0 Hard they report 79.1. Those are vendor numbers, and the lead on Terminal-Bench is narrow, so treat them as a strong invitation to test, not as a settled crown. Still: a 65B dense model sitting next to models with hundreds of billions of total parameters is the interesting part.

Why dense matters here

Ricursive built Ripple to power its own AI chip-design platform, then released a general-purpose preview. The architecture is deliberately simple: a thin-and-deep Transformer with 80 layers and d_model 5120, designed for NVIDIA hardware. They say they kept about 40% MFU through pretraining and posttraining with plain sharding, no custom kernels and no quantization during training. Full R&D cost, including ladders, data, evals and the final runs: $32 million, which they frame as roughly one-tenth of the previous U.S. open-weight coding SOTA.

They also report reaching that Terminal-Bench score with about 1.4 million RL rollouts, far fewer than the prior leader's published figure. Again: different setups, different accounting. The honest takeaway is the direction of travel. Careful model laddering from an Olmo 3 baseline, synthetic posttraining data they generated themselves, and a dense envelope that is easier to train and to serve.

Why I care

For agents and coding tools, total parameter count is less useful than what you can actually run. Ripple ships BF16 and NVFP4 instruct checkpoints plus a BF16 base, and Ricursive points at local boxes like NVIDIA DGX Spark. Hosted endpoints are already listed on Hugging Face, Together AI, GMI Cloud, OpenRouter and Fireworks. Apache 2.0 means you can adapt it commercially without a special model-license deal.

I will not pretend every benchmark is settled. On DeepSWE 1.1, Meta's Muse Spark Open still leads in their own table. Context advertised at up to 1M tokens may not match every hoster's live limit. What is clear is the pattern: smaller dense open weights, trained with discipline, competing on agentic coding. That is exactly the kind of release that helps people building real tools, not just demos.

Sources

Would you rather ship a coding agent on a 65B dense open model you can host yourself, or keep waiting for the next trillion-parameter MoE?

Reacties

Een reactie posten

Populaire posts van deze blog

Meta just shared the full recipe for a 30B agent model you can run at home 🧠

When Meta released Muse Glimmer in August, the pitch was simple: an open-weight, Apache 2.0 model built for AI agents that run on your own Mac or PC. What was missing was the "how". On 19 October, Meta Superintelligence Labs published the full technical report, and for anyone who cares about distillation and local agents, it is one of the most useful reads in a while. What the model is Muse Glimmer is a dense model of about 29.6 billion parameters, including a 1.8B perception encoder, so it reads text and images and writes text. It has a 128K+ context window and was trained on data from more than a hundred languages. The design goal was strict: everything has to fit in 32 GB of GPU memory, and a smaller 4-bit variant targets 24 GB cards. The parts I found most interesting Dense beat mixture-of-experts here. That surprised me. The team trained scaling ladders for a dense model and two MoE designs with the same memory footprint. Under a fixed memory ceiling and their compu...
Meer lezen

AI that can actually "see" a crystal 🔬

Most of the solid stuff around us, from battery materials to chips, is made of crystals: atoms stacked in neat, repeating 3D patterns. Large language models are great at reading papers about those materials, but they have never really looked at the structures themselves. A team at MIT just showed a clever way around that, and the trick is one I keep coming back to: good synthetic data. What they built The model is called CLaC, short for Contrastive Language-Crystal model, and it is described in a paper in Nature Communications by Yang Jeong Park, Mayank Kumaran, Chia-Wei Hsu, Elsa Olivetti and Ju Li. MIT News covered it on 18 October 2026. If you know CLIP, the model that taught image generators to connect pictures and words, you already get the idea. CLaC has two branches: a graph neural network that reads the 3D crystal structure, and a text encoder that reads a description. Contrastive training pulls matching structure-text pairs together in one shared space and pushes mismatche...
Meer lezen

A 35B model on a Mac mini, using just 2.9 GiB of memory 🚀

There's something special about running a capable language model on hardware you own: no API bill, no data leaving the house, just a small box on your desk doing the thinking. A paper that's currently #2 on Hugging Face's trending list takes that idea further than I expected: it runs a 35-billion-parameter model on a Mac mini at about 20 tokens per second while using under 3 GiB of memory. The trick: leave the experts on the SSD The paper is called The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction , and the system is named Edge0 . It builds on a Mixture-of-Experts (MoE) model, Qwen3.6-35B-A3B. In an MoE, each token only uses a few "experts" per layer (here 4 out of 256 in each of 40 layers), so most of the weights sit idle at any given moment. Edge0 takes that literally. The expert weights (quantized to int4) stay on the SSD and are streamed in only when needed. The clever part is a small prerouter : a tiny trained he...
Meer lezen