OpenAI unveiled benchmarks for its custom inference ASIC Jalapeño, co-designed with Broadcom, claiming 1.5–1.9x better performance per watt and up to 3.6x lower latency versus Nvidia's GB200/GB300 on …
A new arXiv preprint (2608.23809v1) investigates whether multilingual LLMs use shared or language-specific features for mathematical reasoning, introducing the Geometry-Invariant Sparse Autoencoder (G…
A developer created Declaude, a tool that converts Claude's writing style into plain English, after spending excessive time and tokens trying to get Claude to avoid its characteristic 'claude-speak' w…
Alibaba's Qwen team released Qwen3.8-Flash-Next on 26 August, a mixture-of-experts model that activates only 6 billion of its 125 billion parameters per token and is billed as an architecture preview …
A new reference from Artificial Analysis, based on vendor documentation fetched August 26, 2026, finds that image and video input billing across AI APIs is highly inconsistent, with the same 1024×1024…
Atomic Chat, a new application, enables users to run open-weight LLMs from Hugging Face locally on Windows, macOS, and Linux, and exposes them through an OpenAI-compatible API for use with coding agen…
Perplexity launched Portable Computer, a fully local version of its agentic Computer platform, on Tuesday, developed with Nvidia, running on Nvidia DGX Spark and Linux machines with RTX GPUs. The syst…
Alibaba Group Holding Ltd. co-founder Jack Ma spent over HK$600 million (approximately $76.5 million) on Hong Kong-listed shares over consecutive trading days, joining Chairman Joe Tsai and CEO Eddie …
Chinese developers including DeepSeek, Alibaba's Qwen, and Moonshot's Kimi are releasing increasingly capable open-weight AI models at a fraction of the cost charged by American frontier laboratories,…
KVBoost, a chunk-level key-value cache reuse system for HuggingFace-compatible decoder models, reduces time-to-first-token by 4.49x (142.4 ms vs. 639.1 ms) on Qwen/Qwen2.5-3B over 1,000 bug-localizati…
XAI's Grok Voice Think Fast 2.0, launched on July 29, 2026, ranked first on the Artificial Analysis Speech-to-Speech Index with a score of 79.0%, including 97% for speech reasoning and 94.7% task succ…
Thomson Reuters launched its first AI model, Thomson-1, on Monday, built on Snowdon, a model created by realigning Alibaba's open-source Qwen model, to reduce reliance on Anthropic's Claude and cut co…
Thomson Reuters is launching its own language model, Thomson, built on Alibaba's Qwen, at a cost of about $40 million over two years, aiming to own its AI instead of renting from OpenAI or Anthropic. …
A report by security research group Crimson Flare warns that popular open-weight AI models can be easily modified to bypass safety measures, potentially enabling terrorists to build explosives and che…
Alibaba Group Holding Ltd. reported a 75% drop in net income for the quarter ended June 30, 2026, and announced a HK$80bn ($10.2bn) share placement in Hong Kong to fund AI investment, the largest prim…
Alibaba Group Holding Ltd. priced a placement of 710 million new Hong Kong-listed shares at HK$112.70 each, raising HK$80 billion (US$10.21 billion), with the transaction expected to close on Aug. 26.…
Qwen, Alibaba's AI team, has released Qwen-Image-2-Pro, a 20-billion-parameter text-to-image model now available on Replicate. The model excels at rendering complex text, particularly Chinese logograp…
S1 Mini, a fine-tune of Qwen 0.6 billion, cleans messy speech-to-text output locally on under 2GB VRAM, consuming just over 1.4GB in testing. It strips filler words, resolves self-corrections, and for…
Alibaba Group Holding Ltd. cut its share repurchases by roughly 80% in the June 2026 quarter, buying back 13.4 million ordinary shares (1.7 million ADSs) for US$162 million versus 56 million ordinary …
Anthropic's Fable model, released the same week as GLM 5.2, costs roughly nine times more than GLM 5.2, prompting agentic coders to adopt cheaper alternatives for routine coding tasks. The high price …