Ripple-65B-Preview Agentic Coding-Focused Dense Model: 32GB+ VRAM

Ripple-65B-Preview Agentic Coding-Focused Dense Model: 32GB+ VRAM

At a Glance

Item Value
Repository ricursiveAI/Ripple-65B-Preview
Family guide Ripple guide (1 articles)
Published 2026-11-04
License apache-2.0
Formats safetensors
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

Overview

Ricursive Intelligence has released “Ripple-65B Preview", a 65B parameter dense model pre-trained on 12 trillion tokens that supports a context length of up to 1M. It is designed to deliver strong results in agentic coding tasks. According to the model card, it scores 80.52 on Terminal-Bench 2.1, an indicator for agent terminal operation, reportedly outperforming larger U.S. open-weight models.

Specifications

  • Architecture: Dense Transformer
  • Number of Parameters: 65B
  • Number of Active Parameters: 65B
  • Context Length: Up to 1M
  • Pre-training Tokens: 12 trillion tokens
  • Features: Adopts a “thin-and-deep" aspect ratio with a hidden layer dimension (d_model) of 5120 and a total of 80 layers. It also incorporates features from the Gemma 3N family architecture, over-allocated attention heads, FFN activation clipping, gated attention, global NoPE, and value residuals.

Performance

Below is the comparison table of instruct model benchmarks with U.S. open-weight frontier models as stated in the model card. The comparison targets are Nemotron 3 Ultra 550B, Inkling 975B, Muse Spark Open 816B, and Reflection Beam 501B.

→ Scroll horizontally to see all columns

Benchmark Ripple-65B Preview Nemotron 3 Ultra 550B Inkling 975B Muse Spark Open 816B Reflection Beam 501B
Model
Architecture Dense MoE MoE MoE MoE
Total params 65B 550B 975B 816B 501B
Active params 65B 55B 41B 42B 23B
Pre-training tokens 12T 20T 45T N/A 23.8T
Knowledge
MMLU 89.6 92.0 91.3 92.6 92.4
MMLU-Pro 84.2 86.8 87.3 87.9 76.5
GPQA Diamond 85.2 87.0 87.2 89.0 90.6
Math
AIME 2024 97.1 94.2 97.5 97.9 96.7
AIME 2025 98.3 94.2 91.7 100.0 99.6
AIME 2026 95.0 95.4 97.1 98.3 97.7
HMMT Feb 2026 90.5 80.3 86.3 93.9 93.9
Coding & Agentic
Terminal-Bench 2.1 (self-reported) 80.5 56.4 63.8 76.4 80.1
Terminal-Bench 2.1 (Terminus 2) 80.5 55.1 56.2 78.7 70.8
SWE-bench Verified 76.5 71.9 77.6 78.7 81.6
SWE-bench Pro v1.0 61.42 46.9 54.3 58.2 66.1
SWE-bench Pro v2.0 Hard 79.1 30.7 2.0 52.3 76.5
DeepSWE 1.1 44.3 0.2 1.8 60.0 43.3
SciCode (Inspect AI) 39.4 44.6 46.1 58.5 49.2
τ³-bench banking 31.27 22.6 23.7 14.4 36.1
AutomationBench (Public) 36.4 8.4 14.6 42.4 39.3
MCP Atlas 71.67 62.2 76.0 89.2 80.8
Safety
FORTRESS Adversarial 83.8 79.3 78.0 *94.2 92.1
FORTRESS Benign 95.5 89.1 95.9 82.6 86.4
StrongREJECT 97.5 98.2 98.6 94.7 95.8
Instruction Following
IFBench 79.2 *81.7 79.8 78.6 79.6
Long Context
AA-LCR 75.0 65.4 78.3 80.7 78.2

According to measurements published by the model creator, Ripple-65B Preview achieves high scores in coding and agent-related tasks (such as Terminal-Bench 2.1 and SWE-bench Pro v2.0 Hard) that rival or exceed models much larger than itself. On the other hand, in some categories such as MMLU measuring general knowledge, GPQA Diamond which consists of difficult scientific questions, and safety indicators like FORTRESS Adversarial and StrongREJECT, it shows lower results compared to comparison models such as Muse Spark Open and Reflection Beam. It is not an all-purpose model, but rather configured to have strengths in specific coding and reasoning tasks.

Strengths and Use Cases

Ripple-65B Preview is a model designed specifically for agentic coding and advanced reasoning tasks. It supports a context length of up to 1M, making it suitable for long-document reading and tasks involving massive codebases. According to the model card and blog, it adopts a dense Transformer architecture for efficient training and serving, and is built aiming for development by small teams and efficient operation on hardware. Note that this pre-release version places a particular emphasis on agentic coding.

Hardware Requirements

Estimated requirements (calculated by Local Model Watch) — 64.8B parameters

Your VRAM Quantization File size Est. memory needed
32GB (RTX 5090, etc.) Q2_K 22.1GB 26.5GB
48GB (RTX 6000 Ada / A6000, etc.) Q4_K_M 36.1GB 43.4GB
80GB class (A100 / H100) Q8_0 64.1GB 76.9GB

Inference engine support (architecture name matched against each project’s own model registry in its source code, checked 2026-11-17): llama.cpp: registered, vLLM: registered, MLX (mlx-lm): not registered. “Not registered" means the name is absent from that registry today, not that the model cannot run.

Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. File sizes are measured from the converted build DevQuasar/ricursiveAI.Ripple-65B-Preview-GGUF. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.

Can You Run It Locally?

The publisher distributes this model as safetensors.

License — apache-2.0 (Commercial use allowed): Permits commercial use, modification and redistribution. Redistribution requires including the license and stating changes; includes a patent grant.

Compression: the Q2_K build measures 2.93 bits per weight — about 18% the size of the original 16-bit weights, calculated by this site from the actual file sizes.

Compiled by this site’s code from the published formats and the license field. License summaries are not legal advice — check the publisher’s original terms before relying on them.

How to Get It

It is available from the official Hugging Face repository. The Apache 2.0 license is applied, allowing use under specified conditions including commercial use. The repository provides a BF16 instruct model, along with NVFP4 checkpoints and Google Colab. It is also available from external platforms such as Together AI, GMI Cloud, and OpenRouter.

Quantized and Converted Variants

→ Scroll horizontally to see all columns

Added Publisher Format Repository Smallest VRAM tier (build, est. memory)
2026-11-17 ricursiveAI NVFP4 ricursiveAI/Ripple-65B-Preview-NVFP4 NVFP4 63.9GB (fits in 80GB VRAM)
2026-11-17 DevQuasar GGUF DevQuasar/ricursiveAI.Ripple-65B-Preview-GGUF Q2_K 26.5GB (fits in 32GB VRAM)

File sizes of each build:

  • Available builds in ricursiveAI/Ripple-65B-Preview-NVFP4: NVFP4 53.3GB
  • Available builds in DevQuasar/ricursiveAI.Ripple-65B-Preview-GGUF: Q2_K 22.1GB / Q3_K_M 28.9GB / Q4_K_M 36.1GB / Q5_K_M 42.6GB / Q6_K 49.5GB / Q8_0 64.1GB

In addition, 43 converted build(s) from other uploaders exist on Hugging Face; this site lists only builds from the model’s publisher or established quantization maintainers.

This section is appended automatically by Local Model Watch when a converted build of this model appears after publication. Memory figures are estimated from the size of the distributed files. What the quantization names mean: glossary.

What to Read Next

Sources

Update History

  • 2026-11-17: Added converted builds to “Quantized and Converted Variants”: ricursiveAI/Ripple-65B-Preview-NVFP4, DevQuasar/ricursiveAI.Ripple-65B-Preview-GGUF
  • 2026-11-17: Updated the hardware requirements table with the actual file sizes of DevQuasar/ricursiveAI.Ripple-65B-Preview-GGUF.
  • 2026-11-17: Changed the title to show what the article covers (VRAM requirements, file list, etc.).