Newsletter

China AI

Xiaohongshu open-sources dots3-note, its IMO perfect-score LLM

The Frontier

Xiaohongshu open-sources dots3-note, its IMO perfect-score LLM

China's AI week keeps delivering: a social-commerce giant just opened the weights of the first model family to ace the International Mathematical Olympiad, and the country's top foundry is cashing in on AI demand like never before. Xiaohongshu's dots lab open-sourced dots3-note preview, the first public release from the dots3 series — the same family as dots-note-3.0, which last month became the first AI system to score an official perfect 42/42 at the International Mathematical Olympiad.

Edge-LLM maker ModelBest files for IPO at ¥20B

The Arena

Edge-LLM maker ModelBest files for IPO at ¥20B

China's on-device AI champion is heading to public markets: ModelBest, the Beijing startup behind the MiniCPM family of edge models, has formally begun the process of listing on the A-share market — the latest sign that China's model labs are treating public listings, not venture rounds, as the endgame. ModelBest filed its IPO counseling registration with the Beijing bureau of the securities regulator on August 11, with CITIC Securities guiding the process, according to Chinese financial…

Ex-Qwen lead's Pragmatik Labs raises $220M at a $2B valuation

The Arena

Ex-Qwen lead's Pragmatik Labs raises $220M at a $2B valuation

Lin Junyang, the researcher who built Alibaba's Qwen into one of the world's most-used open model families, came out of stealth early Wednesday with a company that has no product, no demo, and a $2 billion price tag. That gap is the story. Lin Junyang, the former technical lead of Alibaba's Qwen family, announced the founding of Pragmatik Labs (语用科技) in Shanghai — backed by a $220 million angel round at a $2 billion valuation, closed before the company shipped anything.

Pony.ai's fourth-gen driverless heavy truck enters mass production

The Arena

Pony.ai's fourth-gen driverless heavy truck enters mass production

China's autonomous-driving leader is taking its Robotaxi playbook to freight, with the fourth-generation driverless heavy truck now rolling off the production line and a concrete fleet target for the next three years. Pony.ai's fourth-generation L4 autonomous electric heavy truck has entered mass production, with the company aiming to deploy 500 to 1,000 of the driverless rigs across China within two to three years. The first Gen-4 Robotruck model has already rolled off the line, and the…

China's jobs squeeze: 53M now deliver or drive

The Everyday

China's jobs squeeze: 53M now deliver or drive

Two stories this morning show AI colliding with the real economy from opposite ends: millions of Chinese workers absorbing manufacturing automation as gig drivers, and investors pouring money into securing the AI systems companies are rushing into production. More than 53 million people in China were working as food-delivery or ride-hailing drivers as of 2025 — up roughly 10 million in two years — according to new Financial Times analysis, as a slumping construction sector and accelerating…

LangYa paper: a single AI model now forecasts the global ocean

The Frontier

LangYa paper: a single AI model now forecasts the global ocean

A Chinese research team just published the blueprint for one of the more ambitious AI-for-science systems in the water: a single neural network that forecasts the state of the entire global ocean up to a week out. The Chinese Academy of Sciences' Institute of Oceanology has published the peer-reviewed framework behind LangYa, a large AI model that generates one-to-seven-day forecasts of the global ocean from a single network — no recursive stepping, no ensembles of specialist sub-models. The…

Qwen open-sources Qwen3.8-2.4T-A95B, its first Max-class model

The Frontier

Qwen open-sources Qwen3.8-2.4T-A95B, its first Max-class model

Wednesday's wire: Alibaba opened the weights on its most powerful model ever, Honor put a robotic arm in a phone and sold 400,000 of them before lunch, and Europe's booksellers started refusing suspicious bulk orders. Alibaba's Qwen team open-released Qwen3.8-2.4T-A95B on Wednesday — the first Qwen-Max-class model to ship as open weights, with 2.4 trillion total parameters and 95 billion active in a 512-expert mixture-of-experts architecture.

Coreset pruning cuts 80% of VLM tokens, keeps accuracy intact

The Frontier

Coreset pruning cuts 80% of VLM tokens, keeps accuracy intact

Vision-language models are drowning in their own pixels, and a Chinese Academy of Sciences team has a training-free fix: cut up to 80% of the visual tokens a multimodal model has to chew through — and lose nothing. The CAS Institute of Automation's Zidong Taichu team has released GMC (Grounded Message Coreset Pruning), a training-free method that cuts up to 80.2% of a vision-language model's visual tokens while retaining 97.78% of full-model capability — and in its most aggressive…

Tencent's AI-driven ads beat Q2 estimates, but profit misses

The Arena

Tencent's AI-driven ads beat Q2 estimates, but profit misses

Tencent beat second-quarter revenue estimates on accelerating game sales and AI-driven advertising, but the results laid bare the tension at the heart of its AI bet: core profit missed expectations, and capital spending jumped 65% quarter over quarter. Tencent posted revenue of 204.78 billion yuan ($30.36 billion) for the quarter ending in June, up 11% year on year and ahead of the 202.17 billion yuan analysts expected.

See all