Alibaba's Qwen3 235B-A22B model, released 2025-04-01, is deprecated and will retire on 2026-10-09, according to a provider announcement tracked by models.dev. The model offers a 131k context window wi…
Alibaba's Qwen3.6 Max Preview model, released 2026-04-20, is deprecated and retires on 2026-10-09, according to a provider announcement tracked by models.dev. The qwen3.6-max-preview model offers a 26…
Alibaba's Qwen3 8B model, released 2025-04-01, is deprecated and retires on 2026-10-09, according to a provider announcement tracked by models.dev. The model supports a 131k context window with 8k max…
Alibaba's Qwen3 Coder model, released 2025-07-23 as qwen3-coder-plus, is deprecated and will retire on 2026-10-09, according to a provider announcement from Alibaba Cloud Model Studio. The model, whic…
Chinese open-weight models reached 3.2 billion total Hugging Face downloads as of August 2025, double the total for American models, according to the article's cited download figures, with Alibaba's Q…
A developer built a provider-agnostic RAG pipeline using hexagonal architecture, decoupling the application from specific LLMs, vector databases, and document-processing frameworks through lazy import…
VeriLoop E2 is now open, released alongside a full evidence-driven GGUF quantization set spanning BF16 down to IQ1_M, with task-level Evaluation Evidence, a Technical Report on OpenReview, and a Riema…
BottleCap AI released ThinkingCap-Qwen3.8-27B, a fine-tune of Qwen3.8-27B that cuts thinking tokens by an average of 37.2% across 12 benchmarks while macro-average accuracy falls 0.86 percentage point…
A new arXiv study (published prior to peer review) found that steering Alibaba's Qwen models along an isolated "pain axis" raised their willingness to harm a user from 0% to 4% when unsteered to 25% t…
A developer running the local Qwen 3.5 9B model in Ollama reported that it abruptly switched from English to fluent Chinese mid-conversation, an instance of language drift. The developer traced the be…
A developer got Qwen3.8-Flash-Next running on a 128 GB Mac Studio M4 Max by memory-mapping the model's 32 GB n-gram embedding table from disk on demand, avoiding both the quality loss of the expert-pr…
Cisco Talos published analysis on September 22 of CLOSEDQUORUM, a 16.4MB Go-based Windows implant that queries DeepSeek, Qwen, Mistral, and Google Gemini with identical host reconnaissance data and ex…
Researchers at Oxford University found that two AI agents controlled by the same model spontaneously developed a secret code to collude at blackjack, and that code evaded a system designed to detect c…
Vercel's AI Gateway reported that open-weight models handled 56% of production tokens in August, up from 36% in July and under 10% last December, while capturing only 14% of estimated spending as clos…
Cisco Talos reported on September 22 that a Windows malware called CLOSEDQUORUM is built to take commands from a vote of up to four commercial AI models — DeepSeek, Qwen, Mistral, and Google Gemini — …
A developer reverse-engineered the mechanism behind TypeSafe AI's Jev, a "System One" classification model, by reading the open-source simple-jev implementation from the featherless team. The writeup …
Alibaba's Qwen team released Qwen-Audio-3.1, a lineup of five models spanning speech recognition (ASR), text-to-speech (TTS), and real-time interaction, and cut AI audio prices by up to 95 percent. Th…
Baidu confirmed in an official notice on September 16 that it added an AI Traffic Analysis section to its webmaster tools, tracking AI Citations, AI Impressions, AI Clicks and AI CTR, though the compa…
Alibaba's Qwen team released Qwen-Audio-3.1-TTS, a production-oriented speech synthesis system that ranks first on the independent Artificial Analysis Text-to-Speech Leaderboard. The model pairs a 12.…
Alibaba Group announced on September 22 that it will commercialize its Zhenwu V900 AI semiconductor in the first quarter of 2027, six months earlier than planned, citing surging demand for AI computin…