Alibaba names Dayiheng Liu head of its Qwen LLM team

Share
Alibaba names Dayiheng Liu head of its Qwen LLM team

Alibaba finally has a name on the door of its flagship model line — the day before that name takes the Apsara main stage.

Alibaba has appointed Dayiheng Liu, one of its senior AI researchers, to lead the Qwen large language model team, ending months of ambiguity about who runs the company's most strategically important product. The Information reported the appointment on Monday, citing two employees with knowledge of the matter, and said it brings clarity to the Qwen team's leadership after several rounds of reorganization earlier this year. Chinese financial outlets carried it within hours, and the timing is the tell: Alibaba's Apsara Conference opens in Hangzhou on Tuesday, and the conference site already lists Liu's title as head of the Qwen LLM project under Alibaba's ATH business group and its Token Foundry unit — with his portrait placed directly after chairman Joe Tsai and CEO Eddie Wu on the speaker page. He speaks at 10:15 local time Tuesday, immediately after Wu's opening keynote, on a session titled "Qwen: toward real-world agents."

The seat has been effectively empty since March. Junyang Lin, Qwen's tech lead and the public face of China's most successful open-weight model family, announced his exit on X on March 3 — days after his team shipped four small Qwen3.5 models — and Alibaba's Hong Kong shares slid as much as 5.3% on the news. Alibaba Cloud CTO Zhou Jingren was given interim charge of the model line, and CEO Eddie Wu announced a foundation-model support group in an internal email. That arrangement was always described as temporary. Naming Liu closes it.

Liu is a researcher's researcher rather than a product executive. He holds a PhD from Sichuan University, was part of Huawei's inaugural "Genius Youth" cohort in 2020, and has been one of Qwen's lead foundation-model engineers — a named author on the Qwen3 technical report and on Qwen2-VL. His research record sits in pretraining, reasoning and natural language generation, which is a meaningful signal about where Alibaba thinks Qwen's next increment comes from.

That signal matters because Alibaba has been redrawing the boundary between open and commercial while the leadership question hung open. Qwen is the largest open-weight family in the world by downloads — we covered the milestone when Qwen passed 3 billion downloads — and the company open-sourced a 2.4-trillion-parameter Max-class model in August. Last week it shipped Qwen-Image-2.1 under a research-only licence, a sharp turn from the Apache 2.0 terms its image models carried before. The company has not publicly confirmed Liu's appointment, and the sourcing is two employees rather than an announcement, so treat the title as reported until Alibaba says it.

What to watch: Liu's Tuesday keynote is the first real test of the appointment. If Apsara brings a new flagship model or a concrete agent platform — and if the licence terms on the language side start to look like the image side's — the researcher in the chair will have answered the question the March shake-up left open. Meanwhile Lin Junyang is not idle: his startup Pragmatik Labs raised at a $2 billion valuation this summer, after he left Qwen.

Does promoting an open-weights researcher over a product executive tell you anything about where Alibaba's model strategy is headed — or is it just org-chart cleanup? Tell us in the comments.

Read more

Byteification turns Qwen3 and Llama 3 into byte-level models

Byteification turns Qwen3 and Llama 3 into byte-level models

The tokenizer is the last piece of the language-model pipeline that was never learned end to end — and a Nature paper out this week shows it can be swapped out for less than one percent of a pretraining budget. Researchers have retrofitted four open subword language models into true byte-level models using less than 1% of a typical pretraining budget, and the result — published 7 October in Nature — closes the performance gap that has kept byte-level models a niche curiosity for nearly a decad

OpenAI's dots go mobile and can start your Codex work

OpenAI's dots go mobile and can start your Codex work

One product update and two benchmark papers this hour, all circling the same question: how much of the work can you actually hand over, and what happens to the part you didn't? OpenAI's dots — the always-on agents that get their own cloud computer — can now be created and configured entirely from the ChatGPT app on iOS and Android, and a dot can start Codex work or pick up an existing Codex thread on your behalf. The October 9 release, billed as the first of a fresh run of updates for dots, mo

Coding agents as a hospital: Cockroach's five-month results

Coding agents as a hospital: Cockroach's five-month results

The agentic-coding debate keeps being framed as speed versus trust. Cockroach Labs just published the most detailed trust-first ledger yet — five months of numbers, including the parts that didn't work. Cockroach Labs ran its codebase like a teaching hospital for five months and says the pipeline merged well over a million lines of code with seven total reverts. In the setup the company calls MOLT Sinai, GitHub issues are patients, merging is discharge, and the humans in charge are Chiefs of M

Today in AI — October 10, 2026

Today in AI — October 10, 2026

A day where the boring machinery took center stage: proof checkers, order books, warehouses and the humble text message — plus fresh arguments about who gets to declare any of it safe. Models & Research * Mathematicians get a reliability primer for the Lean Theorem Prover. A guest post by Thomas Hales on Terence Tao's blog walks through what Lean actually guarantees — and devotes itself to the "Summer of Soundness Bugs," the string of kernel bugs found this summer that let false proofs thr