By combining native vision, audio, and reasoning in a compact model, Gemma provides a compelling platform for local AI agents ...

Perplexity's post-trained GLM 5.2 preview pairs a cheaper open-source base with a frontier advisor—and it's already live in ...

Open-source OCR from Baidu eliminates the GPU memory wall that limits long-document parsing. Unlimited OCR uses a constant KV ...

Context windows are becoming a computational bottleneck. The longer an agent runs, the more tokens accumulate from retrieved documents, reasoning traces and conversation history, and the more memory ...

Gemma 4 12B is a 12-billion-parameter decoder-only transformer. It handles text, images, audio, and video natively. There are no separate vision or audio encoders. The decoder uses the same structure ...

We introduce OneCAT, a unified multimodal model that seamlessly integrates understanding, generation, and editing within a novel, pure decoder-only transformer architecture. Our framework uniquely ...

Abstract: Code search is essential for code reuse, allowing developers to efficiently locate relevant code snippets. The advent of powerful decoder-only Large Language Models (LLMs) has revolutionized ...

An extension of the paper with additional results can be found in the provided button above (ExtendedPaper). This repository builds upon the original T-DEED implementation to evaluate the model across ...

Chinese outfit Zhipu AI claims it trained a new model entirely using Huawei hardware, and that it’s the first company to build an advanced model entirely on Chinese hardware. Zhipu, which styles ...

A screenshot of Mu performing real-time question answering. Image: Windows YouTube channel The Mu small language model enables an AI agent to take action on hundreds ...

A new study links layer-time dynamics in Transformer models with real-time human processing. The findings suggest that AI models may not only reach similar outputs as humans but could also follow ...