Alibaba's Qwen team released Qwen-AgentWorld, a family of language world models trained to simulate agentic environments by predicting next states as text, outperforming GPT-5.4 and Claude Opus 4.8 on…
China reaffirmed its support for the United Nations as the primary platform for global AI governance and strongly advocated for open-source AI at the UN’s first Global Dialogue on AI Governance in Gen…
A CTO evaluated four Chinese LLM families—DeepSeek, Qwen, Kimi, and GLM—for production use at scale, finding that DeepSeek V4 Flash offers the best overall value at $0.25/M tokens, while Kimi K2.5 exc…
Two of China's largest AI model platforms, Doubao and Qwen, tightened restrictions on custom agents last week, delisting or throttling third-party agents and narrowing creation permissions. The develo…
A developer building a realtime note-taking app found that Apple's on-device Foundation Models failed to process a 27-minute conversation, while a smaller Qwen model succeeded using a hierarchical map…
A developer spent three weeks comparing AI API prices and found that the cheapest viable models are Apache 2.0 and MIT licensed models from China, not the well-known proprietary names. The analysis re…
A study comparing aligned and refusal-ablated (abliterated) large language models from the Gemma and Qwen families found that abliterated models outperformed aligned models in vulnerability analysis t…
A developer building a realtime note-taking app found that Apple Intelligence's on-device language model failed to process a 27-minute conversation, forcing a hierarchical map-reduce approach to summa…
Researchers introduced BaFCo, a benchmark dataset for Bangla form comprehension comprising 200 multi-page government forms from sectors like agriculture and banking. Evaluations of multimodal large la…
A developer tested four Chinese AI models—DeepSeek, Qwen, Kimi, and GLM—across coding, reasoning, creative writing, and Mandarin tasks. DeepSeek V4 Flash emerged as a top pick for its balance of perfo…
A developer replaced Claude Code with open-source coding agent OpenCode for two weeks, finding it 78% slower but keeping it due to its model-agnostic design that unifies local and hosted AI models. Th…
Hugging Face announced that the transformers vLLM modeling backend now matches or exceeds native vLLM throughput for many LLM architectures, allowing model authors to run their transformers implementa…
Relying on AI APIs creates dependency risks, as models can change or be deprecated without notice. It advocates for owning open-weight models on local hardware to ensure control and stability, calling…
China's Ministry of Commerce has held talks with Alibaba, ByteDance, and Z.ai about restricting overseas access to the country's most advanced AI models, including unreleased ones, according to Reuter…
China is considering new export controls on frontier AI models, including both proprietary and open-weight systems, to protect domestic capabilities amid intensifying competition with the US. Official…
A developer benchmarked four Chinese large language models—DeepSeek, Qwen, Kimi, and GLM—over six weeks using 200 production prompts. DeepSeek V4 Flash offered the best cost-performance at $0.25 per m…
A developer migrated an AI application from Gemini and Google Cloud to Qwen and Alibaba Cloud for the Global AI Hackathon Series with Qwen Cloud. The project, CloudPort Agent, documents migration trap…
Crusoe launched Serverless Fine-Tuning and Self-Serve Deployments for its Intelligence Foundry, enabling organizations to fine-tune and deploy open AI models with granular control and cost efficiency.…
A developer at CommandCodeAI built a tool-input repair layer that enabled DeepSeek V4 Pro to outperform Opus 4.7 on internal evals after analyzing billions of tokens. The fix addressed four common fai…
China's Ministry of Commerce has held meetings with Alibaba, ByteDance, and startup Z.ai to discuss limiting overseas access to the country's most capable AI models, including open-weight systems like…