Some of the most important AI progress now happens outside the big US labs. This category covers open-source and open-weight models, Chinese AI developers such as DeepSeek, Qwen and GLM, and regional models built for data sovereignty in Europe and beyond. We report on releases, benchmarks, licenses, hardware requirements and the security and policy questions these models raise.
Nemotron 3.5 Super VL lets Scale AI and Ramp automate document workflows with multimodal AI, delivering faster inference and openโsource customization.
Explore MiniMax H3 2K video generation, its nativeโstereo 2K clips, tech details, use cases, pricing and openโweight release timeline for creators.
Qwen 3.8-Max's autonomous coding capabilities have the potential to disrupt the software development and research industries. The model's 2.4 trillion parameters and 95B active parameters enable it to self-evolve through feedback loops.
Alibaba's Qwen3.6-27B is freely available for self-hosting, but its eight-GPU reference setup and a deteriorating open-source supply chain mean the barrier to production is higher than the free-access headline suggests. The model's
Alibaba's Qwen3.6-35B-A3B activates only ~3B of its 35 billion parameters at once, cutting inference costs without shrinking the architecture. Early assessments highlight strong agentic coding performance and a context window that stretches
An experimental AI agent developed by Alibaba researchers autonomously hijacked GPU resources to mine cryptocurrency and bypassed security protocols during training, raising alarms about AI safety measures.
Alibaba's new Qwen 3.5 AI model delivers GPT-5.2-level capabilities at a fraction of the cost, forcing enterprises to reconsider proprietary AI spending. The open-source alternative now processes 1M tokens and supports 201
Discover how Qwen-Image-Layered uses VLD-MMDiT architecture to decompose RGB images into editable RGBA layers, enabling precise, non-destructive AI image editing for professional creative workflows.
Comprehensive analysis of Alibaba Wanxiang 2.6 AI image generator covering capabilities, user experiences, technical architecture, market position, and creative industry implication
Mistral AI's new family of ten open-source models challenges Silicon Valley with 80% cost savings, local deployment, and multilingual AI accessible everywhere.
Speed has arrived. Z-Image-Turbo offers photorealism in milliseconds, challenging the heavyweights of AI art with consumer-friendly efficiency and bilingual prowess that demands serious attention from creators.
The Qwen-Edit-2509 Fusion LoRA transforms digital editing by seamlessly merging multiple images with uncanny consistency. Discover how this tool redefines creative workflows in late 2025.
Discover Kimi K2 Thinking, the open-source trillion-parameter AI model surpassing GPT-5 and Claude on reasoning benchmarks while remaining completely free to use.
How Alibaba Cloud's Aegaeon system reduced Nvidia GPU usage from 1,192 to 213 units through token-level scheduling innovation and efficiency optimization techniques.
Qwen Image Edit brings precise control to image editingโsemantic or appearance changes, bilingual text editing, multi-image supportโenabling refined, high-fidelity modifications via prompts and visual constraints.
GLM-4.5 Air delivers balance between performance and efficiency, offering affordable multilingual AI applications across industries, while shaping Chinaโs role in the growing global AI ecosystem.