Digital Event Horizon
Inkling 2 Mini, a cutting-edge looped transformer model, has been released by Thinking Machines Lab. This model boasts impressive performance and unique features that set it apart from its predecessors. With its ability to visit certain layers multiple times, Inkling 2 Mini increases its capacity and outperforms other models in various benchmarks. This release marks a significant milestone in the development of transformer models and offers a versatile solution for a wide range of applications.
INKLING 2 MINI is a cutting-edge looped transformer model designed for local AI setups. It boasts 26 billion physical parameters and 3 billion active parameters per token, enabling out-of-the-box performance. The model supports both image and text inputs, making it ideal for multimodal processing applications. INKLING 2 MINI has been tested in demos and vibe tests, showcasing its potential for real-world applications. The model outperforms its competitors in 32 benchmarks, solidifying its position as a top-tier model.
In a significant update to the Hugging Face ecosystem, Thinking Machines Lab has released Inkling 2 Mini, a cutting-edge looped transformer model designed for local AI setups. This model boasts impressive performance and unique features that set it apart from its predecessors.
According to the release notes, Inkling 2 Mini is a multimodal Mixture-of-Experts model with approximately 26 billion physical parameters and 3 billion active parameters per token. Its architecture is characterized by a looped transformer design, which allows it to visit certain layers multiple times, effectively increasing its capacity. This innovative approach enables Inkling 2 Mini to outperform other models in various benchmarks.
One of the standout features of Inkling 2 Mini is its support for both image and text inputs. This makes it an ideal choice for applications that require multimodal processing, such as image-text-to-text translation or image captioning. The model also includes a DSpark drafter, which provides an additional speed-up at a small memory cost.
In addition to its technical capabilities, Inkling 2 Mini has been tested in various demos and vibe tests, showcasing its potential for real-world applications. These tests include generating 3D shaders, recreating famous paintings, and even building a Trip Split app. The results demonstrate the model's ability to learn and adapt to complex tasks.
Inkling 2 Mini has also been benchmarked against other models, including Gemma 4, Muse Glimmer, Qwen3.8, and Nemotron 3.5 Super VL. The results show that Inkling 2 Mini outperforms its competitors in 32 benchmarks, solidifying its position as a top-tier model.
The release of Inkling 2 Mini is accompanied by a detailed technical overview, including information on its architecture, inference support, and deployment options. The model is available on GitHub and can be easily integrated into various applications using the Hugging Face ecosystem.
In conclusion, the release of Inkling 2 Mini represents a significant milestone in the development of transformer models. Its innovative architecture, impressive performance, and versatility make it an attractive choice for a wide range of applications.
Related Information:
https://www.digitaleventhorizon.com/articles/New-Benchmarking-Results-and-Release-of-Inkling-2-Mini-a-Cutting-edge-Looped-Transformer-Model-deh.shtml
https://huggingface.co/blog/inkling2mini
Published: Fri Nov 20 15:33:43 2026 by llama3.2 3B Q4_K_M