In this categoryLocal AI · 75
- How to Run Linum-AI/pjit-21k Locally: A Text-to-Image Model Trained for $21,000
- How to Run ConwayResearch/Underdog-Woof Locally: A 38M-Parameter CPU Text-to-Speech Model
- Ripple-65B Preview: The 65B Dense Model That Beats 10x Larger Open Models on Agentic Coding
- How to Run milliwattlabs/leokani-tts-7M Locally: A 7M-Parameter CPU Text-to-Speech Model
- SenseNova 6.8 Code2Video: The 27B Model That Writes Code, Not Pixels, to Make Video
- Speridlabs Idol: Turn One Photo into a 3D Gaussian Splat, and How to Run It
- LFM3-Omni-2B: Liquid AI's 2.2B Any-to-Any Model, and How to Run It Locally
- SenseNova-Loop-9B-Preview: A Looped Multimodal Model for Long-Context Work, and How to Run It
- How to Run WeVisOne-VL-4B Locally: Tencent's All-in-One Vision Model
- How to Run Mistral Large 4 (1T-A52B) Locally: The Hardware It Actually Needs
- How to Run Hy-SimulMT-30B-A3B-FP8 Locally: Tencent's Real-Time Speech Translator
- How to Run Chroma1-HD Locally: An Open Base Model for Text-to-Image
- How to Run Nemotron 3.5 Parse Locally: NVIDIA's 630M Document Parsing Model
- How to Run Recursive Flint Locally: a 39.8B Coding Model You Can Actually Fit on a Mac
- How to Run Phonon-40M-en Locally: Gradium's 40M-Parameter Text-to-Speech Model
- How to Run Timestral-209M Locally: Mistral's 209M-Parameter Time Series Forecasting Model
- How to Run Indra OCR Locally: a 1M-Parameter CPU Model That Hits 49 Pages a Second
- How to Run AuroraIMG-6M Locally: AuroraAI-Research's 6.82M-Parameter Text-to-Image Model
- How to Run Maya-1 Locally: a 3.3B-Parameter Text-to-Speech Model with Voice Design
- How to Run Sopro V2 Turbo Locally: a 120M-Parameter Voice Cloning TTS Model
- How to Run Kova TTS 1 Locally: a 1.25B-Parameter Voice Cloning Model for Narration
- How to Run Whistle Locally: Cactus's 16.9MB On-Device Speech Recognition Model
- How to Run Antalia-2 Mini Locally: a 7.6M-Parameter Turkish TTS Model
- How to Run Supra2-IMG Locally: SupraLabs' 104M-Parameter Text-to-Image Model
- How to Run VisionHOPE Locally: PSRben's Self-Modifying Vision Backbone
- Fish Audio S1-Mini: What It Is and How to Actually Try It
- How to Run SupraTTS-0.1-Beta Locally: A Tiny Glow-TTS Voice Model
- How to Run Nemotron 3 Diarization Locally: Streaming Speaker Tagging in 99M Parameters
- How to Run Freya-TTS Locally: A 183M-Parameter Turkish Text-to-Speech Model
- How to Run Audio8 ASR Infinite Locally: Streaming Speech Recognition That Never Stops
- How to Run TeleOCR Locally: One Model for Scanned and Photographed Documents
- How to Run Xing4.0-29B-A4B Locally: A 256K-Context MoE Model for Agent Tasks
- How to Run IndicF5 Locally: Voice Cloning for 11 Indian Languages
- How to Run Fish Audio S2 Pro Locally: Multilingual TTS with Inline Emotion Control
- How to Run ZDTaichu5.0-9B Locally: A 9.79B Vision-Language Model Built for Spatial Reasoning
- How to Run Qwen3-TTS-12Hz-1.7B-VoiceDesign Locally: Voice Design From a Text Description
- How to Run YuE2-3B Locally: Open Source Music Generation That Beats Suno v5 on Benchmarks
- MiniCPM5-2B: Install and Run OpenBMB's 2B Model That Outscores 4B Rivals
- BERT Base Uncased: Specs, How to Run It, and Why It's Still Trending in 2026
- How to Run tencent/AuK Locally: Zero-Shot TTS, Voice Cloning and Speech Editing
- Stable Diffusion 3.5 Medium: What's Different From SDXL and Flux
- Qwen-Image-2512: The Text-to-Image Model That Renders Real Text
- How to Run Irodori-TTS Anime Locally: Japanese Voice Cloning and Voice Design
- How to Run Breeze TTS 2 Locally: Voice Clone, Voice Design and Voice Direction
- How to Run Kokoro-82M Locally: the Fastest Open-Weight TTS Model
- How to Install Ollama on macOSStart
- How to Install Ollama on Windows
- How to Run Llama 3 Locally with Ollama
- How to Pick the Right Local AI Model for Your Hardware
- Best GGUF Models to Run by VRAM Tier (8GB, 12GB, 16GB, 24GB, 48GB)
- Run LLMs in Your Browser With WebGPU: No Install, No Server (WebLLM)
- How to Use Ollama as a Drop-In OpenAI API
- GGUF vs MLX vs NVFP4: Local AI Quantization Formats Explained
- Best GPU for Running AI Locally in 2026Start
- How Much RAM Do You Need for Local AI?
- Mac vs PC for Local AI: Which Should You Choose?
- How to Build a Local AI Workstation on Any Budget
- How to Set Up Local AI on 8GB of VRAM
- How to Set Up Local AI on 12GB of VRAM
- Best Local LLM for 16 GB VRAM: Setup, Quantization and Real Speed
- How to Set Up Local AI on 24GB of VRAM
- Best Local LLM on Mac M4 16 GB: Setup, MLX vs GGUF and Real Speed
- Best Local LLM on Mac M4 Pro 48GB: Setup, Quantization and Real Speed
Ripple-65B Preview: The 65B Dense Model That Beats 10x Larger Open Models on Agentic Coding
Ripple-65B Preview is a 65 billion parameter dense language model from Ricursive Intelligence, released November 4, 2026 under Apache 2.0. It scores 80.52 on Terminal-Bench 2.1, ahead of several open-weight models many times its size, and ships in BF16 and NVFP4 builds, the NVFP4 checkpoint built to run on a single NVIDIA DGX Spark.
Ripple-65B Preview is a 65 billion parameter dense language model from Ricursive Intelligence, released November 4, 2026 under Apache 2.0. It scores 80.52 on Terminal-Bench 2.1, ahead of several open-weight models many times its size, and ships in BF16 and NVFP4 builds, the NVFP4 checkpoint built to run on a single NVIDIA DGX Spark.
Ricursive Intelligence is a new lab, and Ripple-65B Preview is its first public model. It climbed into the top 20 of Hugging Face's overall trending list within a day of release, which is a fast start for a lab nobody had a model card for a week earlier. The pitch is unusual for 2026: instead of a mixture-of-experts model chasing parameter count, Ricursive trained a dense transformer and is betting that careful architecture and data choices close the gap with models ten times its size.
Written by Priya Raghunathan, local-AI hardware reviewer. I verify parameter counts, licenses, hardware requirements, and every command against the primary source before recommending a setup path.
A new top-20 entrant on Hugging Face's overall trending list
At time of writing the Ripple-65B Preview model page shows 271 likes and 939 downloads. The only linked Space, akhaliq's ZeroGPU demo, has no likes yet, and Ricursive has not published a dedicated GitHub repo for the model itself. The one GitHub link on the model card points to Allen AI's Olmes eval harness, at 407 stars, which Ricursive forked internally to run the base-model benchmarks it couldn't source from other labs' own reporting.
Specs at a glance
| Field | Value |
|---|---|
| Publisher | Ricursive Intelligence (ricursiveAI) |
| Architecture | Dense transformer, custom "Ricursive" model type |
| Total parameters | 64.78B (safetensors total) |
| Pretraining data | 12 trillion tokens, mostly English text and code |
| Context window | Up to 1,048,576 tokens on the full BF16 checkpoint |
| Checkpoints | BF16 instruct, NVFP4 instruct, BF16 base, DFlash2 speculative draft |
| License | Apache 2.0 |
| Release date | November 4, 2026 |
Why dense instead of mixture-of-experts
Every other model Ricursive compares itself against, Nemotron 3 Ultra, Inkling, Muse Spark Open, and Reflection Beam, is a mixture-of-experts design running between 501B and 975B total parameters. Ripple-65B is a plain dense transformer, meaning all 64.78B parameters activate on every token rather than routing to a subset of experts. Ricursive says that choice was deliberate for a first model: it keeps the training and serving stack simple, with no custom kernels and no quantization during training, while a small team of one to five people scaled up the project.
The model also uses what Ricursive calls a thin-and-deep layout, a narrower hidden size of 5,120 stacked across 80 layers, plus a Gemma 3N style attention pattern instead of the Llama-style block their earlier Olmo 3 baseline used. On top of that they added extra attention heads, adjusted the ratio of sliding-window to full attention layers, and layered in a handful of smaller tweaks (FFN activation clipping, gated attention, no positional encoding in some layers, value residuals) aimed specifically at reasoning and multi-step agent tasks rather than raw perplexity.
How it stacks up against much larger models
Ricursive's benchmark table runs Ripple-65B against four larger open-weight models on the same evals, using each lab's own reported numbers where available and its own eval runs for the rest. Terminal-Bench 2.1, a test of multi-step command-line and agentic tasks, is where the size gap is most visible.
| Benchmark | Ripple-65B (65B) | Nemotron 3 Ultra (550B) | Inkling (975B) | Muse Spark Open (816B) | Reflection Beam (501B) |
|---|---|---|---|---|---|
| Terminal-Bench 2.1 (self-reported) | 80.5 | 56.4 | 63.8 | 76.4 | 80.1 |
| SWE-bench Verified | 76.5 | 71.9 | 77.6 | 78.7 | 81.6 |
| SWE-bench Pro v2.0 Hard | 79.1 | 30.7 | 2.0 | 52.3 | 76.5 |
| AIME 2025 | 98.3 | 94.2 | 91.7 | 100.0 | 99.6 |
| MMLU-Pro | 84.2 | 86.8 | 87.3 | 87.9 | 76.5 |
At only 65B total parameters, the model can fit on consumer hardware such as the NVIDIA DGX Spark.
SWE-bench Pro v2.0 Hard is the starkest result in the table. Ripple-65B's 79.1 is closer to the much larger Reflection Beam's 76.5 than any of the mid-size mixture-of-experts models manage, with Inkling scoring a near-total 2.0 on that specific eval. Nemotron 3 Ultra, by contrast, leads most of the plain knowledge benchmarks like MMLU-Pro, which tracks with Ricursive's own explanation: Nemotron leaned heavily on synthetic training data tuned for traditional pretraining evals, a trade that doesn't carry over as well to agentic coding tasks.
Running it: two checkpoints for two very different setups
The BF16 instruct checkpoint's safetensors files total roughly 129.6 GB, more than any single consumer GPU's VRAM. Ricursive's own serving example reflects that: it runs on two NVIDIA B200 GPUs through SGLang 0.5.19, paired with the separate DFlash2 draft model for speculative decoding.
That command runs serve.py from the model repo's inference/server folder, which Ricursive says installs its own dependencies through uv. Defaults are tensor parallel size 2, BF16 weights and KV cache, and a 1,048,576-token context window. The server exposes an OpenAI-compatible chat endpoint on 127.0.0.1:30000, including tool calling with strict JSON schemas.
For a single consumer-grade box, Ricursive points instead to the NVFP4 checkpoint, which keeps attention layers in BF16 and quantizes only the MLP layers, run through a bundled spark.py script meant for a Jupyter notebook on an NVIDIA DGX Spark. The script checks the GPU's reported compute capability and refuses to start on anything other than the DGX Spark's GB10 chip.
from spark import Spark
engine = Spark(
target="ricursiveAI/Ripple-65B-Preview-NVFP4",
draft="ricursiveAI/Ripple-65B-Preview-DFlash2",
)
print(engine.chat("Explain NVFP4 quantization in two sentences."))License and use
Ripple-65B Preview ships under Apache 2.0, so commercial use, modification, and redistribution are allowed without a separate agreement. Ricursive flags this as a Preview release with comparatively light posttraining focused on agentic coding, so expect rougher edges than a model several training cycles further along.
FAQ
What is ricursiveAI/Ripple-65B-Preview?
It is a 65 billion parameter dense language model, the first public release from the lab Ricursive Intelligence, trained on 12 trillion tokens and released under Apache 2.0 on November 4, 2026.
How many parameters does it have?
64.78 billion parameters total, per the model's safetensors total on Hugging Face, and every parameter activates on every token since it's a dense model rather than mixture-of-experts.
Can I run Ripple-65B on a single consumer GPU?
Not the BF16 checkpoint; its weights alone total about 129.6 GB. Ricursive's own BF16 example uses two NVIDIA B200 GPUs. For local use on a single machine, Ricursive publishes an NVFP4 checkpoint built to run on one NVIDIA DGX Spark, with a smaller 131,072-token context window.
Why does it need trust_remote_code?
Ripple-65B uses a custom architecture, registered as model type "ricursive" with its own modeling and configuration Python files in the repo, rather than an architecture already built into transformers. Loaders need to either enable trust_remote_code or register the model class directly, as Ricursive's own SGLang scripts do.
Is it free for commercial use?
Yes. Apache 2.0 permits commercial use, modification, and redistribution without additional permission.
Where to go from here
- Try it without installing anything: akhaliq's ZeroGPU Space runs the model on Hugging Face's shared GPUs.
- Read the model card for the base checkpoint and full benchmark tables: the Ripple-65B Preview page on Hugging Face.
- Compare it with other local models: the local AI model directory has specs side by side.
- Browse other recent how-to guides: the how-to index lists them newest first.
Local models run better with more VRAM. CompareRTX GPUs on Amazonbefore you upgrade.(affiliate link. We may earn a commission at no extra cost. Disclosure)
Watch related tutorials
10:05
16:20
12:55
11:35
9:42
10:30Weekly local AI drops
New models, what runs on your hardware, and the guides to set them up. One email a week, unsubscribe any time.