open-source

Baidu has open-sourced its ERNIE 4.5 AI models, featuring multi-modal Mixture-of-Experts with enhanced orthogonality. Modong Space, China’s national computing platform, is the first to integrate these advanced models. ERNIE 4.5 optimizes training and inference for improved text generation, image comprehension, and multi-modal reasoning, catering to diverse real-world applications.

Baidu has released its ERNIE 4.5 open-source large language model series, featuring ten models including Mixture-of-Experts (MoE) variants. These models offer fully open-sourced pre-training weights and inference code, accessible via platforms like Hugging Face. The series boasts innovative heterogeneous MoE architecture for multimodal capabilities and achieves state-of-the-art performance on various benchmarks, outperforming competitors in text and multimodal tasks. The models are distributed under the Apache 2.0 license, promoting both academic and commercial use.

Huawei has open-sourced its Pangu 7B dense and Pangu-Pro MoE 72B large language models, along with Ascend-based inference technology. This move supports Huawei’s Ascend ecosystem strategy, aiming to accelerate AI research and application. The Pangu-Pro MoE 72B model shows strong performance, ranking highly on benchmarks for models under 100 billion parameters.

Moonshot AI launched Kimi-Researcher, its first autonomous AI agent, currently in beta. Built on end-to-end agentic RL, Kimi-Researcher surpasses leading models like Claude 4 Opus and Gemini 2.5 Pro in internal tests, demonstrating strong autonomy and zero-structure adaptability. The agent independently manages research tasks, navigates conflicting information, and prioritizes accurate results. Moonshot AI plans to open-source key components to further accelerate advancements in agentic RL.

HDC 2025 in Sanya showcased advancements in the HarmonyOS ecosystem, specifically focusing on streamlining app development. Huawei spearheaded the unveiling of the HarmonyOS Application Open Source Component Technical Map, a collaborative effort with partners to create a comprehensive library of open-source components. This initiative aims to reduce redundant coding, accelerate development cycles, and foster a thriving environment for developers, featuring 1115 open-source libraries and 7 cross-platform frameworks. The event included presentations from industry leaders demonstrating successful framework integrations and the growth of the HarmonyOS platform.

DeepSeek R1-0528 surpassed Claude Opus 4 in web programming, showcasing strong coding capabilities, especially with detailed prompts. Though performing well in web app development, it faltered with a Tetris game. R1 is also the top-performing open-source text model but has been recently surpassed by Kimi-Dev in open-source code.

Moonshot AI introduced Kimi-Dev-72B, an open-source code LLM designed for software engineering. The 72-billion parameter model tops the SWE-bench Verified benchmark, surpassing larger models like DeepSeek-R1. Kimi-Dev-72B utilizes RLHF to autonomously repair code within Docker. Key features include BugFixer, TestWriter, mid-stage training, and test-time self-play. The model is available on Hugging Face and GitHub.

Alibaba open-sourced its modular Wan2.1-VACE video generation model (1.3B/14B parameters), featuring multimodal input support (text, images, video clips) and Lego-like customizable modules. The lightweight 1.3B version runs on consumer GPUs, democratizing AI video creation. Released on GitHub/Hugging Face, it has gained 330K+ downloads and 11K+ stars, becoming a leading open-source video generation framework.