1. Qwen3.8-27B Released, Tops ModelScope Popularity Chart Community
Alibaba Cloud’s Tongyi Qianwen team released the Qwen3.8-27B open-source large model on the ModelScope platform. The model debuts with 27 billion parameters and is the latest iteration of the Qwen series. Today it received 245,735 popularity points on ModelScope, making it the most-watched open-source model release on the platform that day. Model weights are now available for download, and developers can directly invoke or deploy the model via ModelScope.
2. Kimi-K3 Model Launches on ModelScope Community
Moonshot AI released the Kimi-K3 model on ModelScope. This model is the latest version of the Kimi series. The model page is now live, supporting online experience and weight downloads. Kimi-K3 continues the Kimi series’ technical approach in long-text processing and reasoning capabilities. Specific parameter counts and technical details are subject to the model card.
3. GLM-5.3 Model Released Community
ZhipuAI released the GLM-5.3 model on the ModelScope platform. This is the latest iteration of the GLM series, and the model page is now online with downloads available. The GLM series is ZhipuAI’s self-developed open-source large model family. This 5.3 update comes several months after the previous release, with improvements including optimized reasoning efficiency and instruction-following capabilities. A detailed technical report has not yet been published.
4. Paul Christiano Joins OpenAI Foundation Board of Directors Community
OpenAI officially announced that Paul Christiano, founder of the Alignment Research Center, has joined the OpenAI Foundation Board of Directors and its Safety and Security Committee. The committee is responsible for governance oversight of OpenAI’s company-wide safety and security practices. Christiano has long-standing research experience in AI alignment and previously worked at NIST (National Institute of Standards and Technology) for many years. OpenAI stated that his addition brings an independent perspective to the foundation’s oversight work, helping to challenge existing assumptions, evaluate safety measures, and strengthen governance mechanisms.
5. DeepSeek V4.1 Flash Released, Another Open-Source Model Iteration Community
DeepSeek released the DeepSeek V4.1 Flash model, with model weights simultaneously uploaded to Hugging Face. V4.1 Flash is positioned as an efficient inference version, optimizing inference speed and resource usage while maintaining the core capabilities of the V4 series. The model follows DeepSeek’s consistent open-source approach, with fully open weights. Developers can directly download and deploy via Hugging Face. The release sparked extensive discussion on Hacker News.
6. OpenAI Discloses Details of Large-Scale Security Defense Operation Community
OpenAI disclosed a large-scale security defense operation: mobilizing over 250 people to cover defense hardening across hundreds of systems. OpenAI stated that its latest network model helped identify several vulnerabilities that might otherwise have gone undetected, and completed vulnerability validation and fix confirmation. OpenAI also published the lessons learned, architecture design, and a practical playbook from the operation, describing how to build a “Defense Factory” — a loop mechanism in which AI agents continuously discover vulnerabilities, validate them, and confirm the effectiveness of fixes.
7. Cohere Launches LLM Serving System Based on Decode Megakernel Community
Cohere released a new-generation LLM text generation serving system, the first complete serving solution built around a decode megakernel. According to data published by Cohere, the system achieves up to a 1.58x performance improvement over vLLM. The system is purpose-built for the North Mini Code model, and the code is fully open source. Cohere stated that this architecture represents a new direction in low-level compute optimization for LLM text generation serving.
8. Mistral Integrates Inference Infrastructure and Open-Source Models, Positioning for European AI Sovereignty Community
Mistral AI announced the integration of inference infrastructure, open-source models, and long-term commitments to provide Europe with the complete technology stack needed to take control of its AI future. In its statement, Mistral positioned itself as a key player in building Europe’s AI infrastructure and unveiled a global roadmap. The company said this move aims to reduce Europe’s external dependence at the AI infrastructure level while offering a reference path for global AI development.
9. Qwen2-VL Released, Multimodal Understanding Capabilities Fully Upgraded Community
The Tongyi Qianwen team officially released Qwen2-VL, the latest vision-language model based on the Qwen2 series. Compared to the previous generation Qwen-VL, Qwen2-VL delivers significant upgrades in multiple areas: it supports image understanding at arbitrary resolutions and aspect ratios, achieves state-of-the-art (SoTA) performance on visual understanding benchmarks such as MathVista, DocVQA, RealWorldQA, and MTVQA, and enhances video content understanding. The model is available with a demo, GitHub code, Hugging Face and ModelScope weight downloads, and API access.
10. Qwen2.5-Coder Released, Another Step Forward for Open-Source Code Models Community
The Tongyi Qianwen team released the next-generation open-source code model Qwen2.5-Coder and officially renamed the CodeQwen series to Qwen-Coder. The series’ previous release, CodeQwen1.5, had already gained widespread community attention, and this update further improves code generation capabilities. The team stated that the name “Coder” better reflects a human-like collaborative positioning, with the vision of making the model a true programming partner. The model is available for download via multiple channels including GitHub, Hugging Face, and ModelScope, and offers an online demo.
Today’s Focus: Dense Open-Source Large Model Releases — Qwen3.8-27B, Kimi-K3, GLM-5.3, and DeepSeek V4.1 Flash Debut on the Same Day, Open-Source Competition Enters a New Phase.