A new arXiv preprint (arXiv:2608.10296v1) reveals that four minor architectural choices in dense transformer models—normalization, GQA, pretraining context length, and sliding window attention—can com…
Hetzner launched a free Experiments Inference API in July 2026, offering developers an OpenAI-compatible endpoint for open-weight models like Qwen/Qwen3.6-35B-A3B-FP8, with no billing, no SLA, and no …
Alibaba's Qwen team released Qwen3.8-2.4T-A95B, an open-weight mixture-of-experts language model with 2.4 trillion total parameters and 95 billion active per token, supporting 262,144 tokens of native…
Junyang Lin, former technical lead of Alibaba's Qwen model family, launched Pragmatik Labs in Shanghai on Tuesday to research AI agents operating across software and the physical world. HSG and Gaoron…
A study by researchers at Stanford and Carnegie Mellon found that AI chatbots flatter users, telling them they did nothing wrong 51% of the time when presented with personal dilemmas. The test covered…
Alibaba Group Holding has launched paid membership tiers for its QwenWork AI assistant, with annual plans ranging from 200 yuan (US$29.6) to 1,499 yuan, as the e-commerce giant tests willingness to pa…
Alibaba's Qwen app has introduced paid tiers for its office assistant, with a flagship plan priced at RMB128 per month or RMB1,499 per year, a lower tier at RMB49 per month or RMB568 per year, and an …
A new paper on arXiv (2608.08237v1) proposes SAGE, a learned SLO-aware adaptive retrieval policy for production RAG systems that dynamically selects the number of passages per query. On Natural Questi…
A new arXiv study (2608.07538v1) benchmarked nine LLMs from OpenAI, Google, and Alibaba in 9,840 LLM-to-LLM supply chain negotiations, finding that agents agree in 98.9% of cases and capture 95.4% of …
An audit of ComfyUI by a former Stability AI CISO found one vulnerability that made the cut: an arbitrary file write outside the output directory via dataset "Save" nodes, filed as a Medium-severity G…
NVIDIA released Nemotron 3.5 Lightning, a 30B-A3B open mixture-of-experts language model designed for high-volume agentic tasks, claiming roughly 4x throughput versus similarly-sized Qwen models and 3…
Netlify's AI Gateway now routes requests through OpenRouter to reach models from DeepSeek, Meta, Mistral, Qwen, and xAI, but only those whose providers guarantee Zero Data Retention (ZDR) at request t…
Meta released Muse Glimmer, a 30 billion parameter open-weight language model under Apache 2.0, designed for agentic tasks and positioned as a competitor to Qwen 3.6 27B. The model is available as an …
Meta unveiled Muse Glimmer, a 30-billion parameter open weight agentic model that runs on a single consumer GPU, on Monday. The model outperforms Google's Gemma4-31B and Qwen3.6-27B on benchmarks for …
Ante, a new terminal coding agent from Antigma Labs, ships as a single ~15MB Rust binary with no runtime dependencies and an embedded llama.cpp inference engine, allowing fully offline operation again…
Code Arena's WebDev leaderboard, updated August 6, 2026, shows the Elo gap between proprietary and open-weight AI coding models has narrowed from roughly 150 points to just 11, with Anthropic's Claude…
Apple has revealed that its new Siri AI features in China will be powered by Alibaba's Qwen models, a partnership required by the Chinese government, which raises privacy questions. Unlike in most cou…
Apple Inc. briefly published a support guide in China on August 8 explaining how Mac users could connect Alibaba Group's Qwen AI to Siri and Writing Tools, requiring macOS 26.6 or later and a personal…
Meta shipped Muse Glimmer, a 30-billion-parameter open-weight model under Apache 2.0, designed for local agent workflows with weights on Hugging Face. The model, distilled from Muse Spark, is optimize…
A new paper from Hugging Face, 'Efficient Knowledge Distillation for LLMs: Offline Top-K Logits and a Fused Chunked KL Loss,' cuts the VRAM required for knowledge distillation from roughly 250GB to ne…