Meituan open-sources LongCat-2.0, a trillion-parameter model it says was trained end-to-end on Chinese chips
Meituan released LongCat-2.0, a 1.6-trillion-parameter open-weight (MIT) model that ran anonymously as 'Owl Alpha' for two months and was, the company says, both trained and served entirely on domestic Chinese AI ASICs with no Nvidia GPUs.
SK Hynix's Nasdaq listing raises $26.5 billion, the largest first-time US listing by a foreign company ever
SK Hynix raised $26.5 billion in a Nasdaq listing that topped Alibaba's record for the largest US IPO ever by a foreign company, priced on the strength of its dominance in the high-bandwidth memory that AI accelerators depend on.
OpenAI reframes ChatGPT from chatbot to 'colleague' with GPT-5.6, ChatGPT Work, and Sites
OpenAI launched a three-part 'colleague' pivot in a single week: the GPT-5.6 model family (Sol, Terra, Luna), ChatGPT Work -- an agent that runs multi-day projects and delivers finished decks and spreadsheets -- and Sites, a chat-driven web-app builder.
OpenAI's GPT-Live handles conversation in real time and delegates the hard thinking to GPT-5.5
OpenAI launched GPT-Live, a full-duplex voice system that decides to speak, listen, pause, or interrupt several times a second -- and hands off any request needing deep reasoning to GPT-5.5 running in the background, keeping the voice fast while the 'brain' stays swappable.
SciReasoner is an AI for science that shows which atoms and bonds its answer rests on
Shanghai AI Lab's SciReasoner turns proteins, molecules, and crystals into discrete tokens the model reasons over out loud, so a scientist can audit which structural evidence its prediction depends on -- and expert reviewers rated its reasoning at least as good as a frontier LLM's in 98% of cases.
Anthropic launches Claude Science, an AI workbench that keeps data in the lab and checks its own citations
Anthropic released Claude Science, a beta workbench that wires Claude into researchers' real tools -- PubMed, Jupyter, HPC clusters -- runs on the lab's own hardware so sensitive data never leaves, and pairs a working agent with a separate reviewer agent that flags and corrects citation and calculation errors.
Analysts warn Nvidia's AI boom is being propped up by a circular financing loop with the 'neoclouds'
An IO Fund analysis details how Nvidia invests billions into GPU-rental 'neoclouds' like CoreWeave that then use that capital to buy Nvidia chips -- a self-reinforcing loop that includes a $6.3 billion Nvidia backstop obligating it to buy CoreWeave's unsold GPU capacity through 2032.
Mesh LLM lets you run models too big for any single machine by splitting them across peers
Mesh LLM, the top project on Hacker News this week, runs models larger than any one machine can hold by partitioning them across networked peers -- layers 0-15 on one node, 16-31 on another -- over a serverless peer-to-peer transport, exposing a standard OpenAI-compatible API on localhost.
Nvidia's 'Puzzle' pipeline shrinks a 120B model to 75B and roughly doubles serving throughput
Nvidia's Nemotron-3-Puzzle compresses a 120-billion-parameter model down to 75 billion using a multi-stage pipeline of pruning, distillation, and quantization -- roughly doubling throughput on a B200 node and lifting single-H100 concurrency at long context from one request to eight.
Cohere open-sources an Arabic speech model that beats Whisper and handles dialects and code-switching
Cohere released an open-source (Apache 2.0) Arabic speech-recognition model that it says is the most accurate open Arabic transcriber to date, cutting word error rate well below OpenAI's Whisper Large V3 and handling dialects and Arabic-English code-switching, not just formal Modern Standard Arabic.
Alibaba's Qwen3Guard flags unsafe AI output token-by-token as it's being generated
Alibaba released Qwen3Guard, its first open safety-guardrail model, whose streaming variant classifies an AI response for safety as each token is generated rather than after the fact -- and adds a 'Controversial' tier between Safe and Unsafe that apps can tune stricter or looser.
LaMem-VLA gives robots a working memory so they stop forgetting what they just did
A new robotics model, LaMem-VLA, fixes the 'short-horizon bias' that makes robots react only to what's in front of them by weaving compressed memories of past actions directly into the model's internal representation -- so it can handle multi-step tasks that require remembering an earlier state.