Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Tencent Expands WorkBuddy Tenfold as Hy4 Calls Surge

Tencent said on November 5 that it has scaled WorkBuddy tenfold to absorb a surge in calls to its Hy4 model, according to QbitAI. The same report describes Hy4 as the leading open-source model on agent capabilities, a ranking the candidate summary does not substantiate with benchmarks.

Published

Tencent said on November 5 that it has expanded WorkBuddy tenfold after a surge in calls to its Hy4 model, according to a report by QbitAI.

A tenfold expansion is a concrete engineering move rather than a routine version bump: inference capacity, concurrency limits and service headroom were all raised together to keep the product responsive under heavier load.

QbitAI's headline also describes Hy4 as the top open-source model on agent capabilities. That ranking is the outlet's characterisation, and the available information does not set out which benchmarks or comparisons it rests on, so it should be read as a claim rather than a measurement.

The timing is the interesting part. Capacity expansion is usually a lagging indicator, so a tenfold jump implies demand moved before the announcement did, and the provider is closing a gap rather than pre-building for one.

Open-weight models are attractive to developers because they can be self-hosted, fine-tuned and cost-controlled, but serving them reliably is a separate problem from releasing them. By making the expansion public, Tencent effectively puts its serving stack into the story too.

It is also a shift in framing. Instead of arguing over benchmark scores, Tencent points at load, and the size of the buildout becomes the credibility behind that argument.

What to watch: the concrete version and parameter count behind Hy4, what WorkBuddy actually does for users, and whether service quality and per-call costs hold after the expansion.

Why it matters

For China's open-model ecosystem this is a public case of proving capability through real usage, making serving capacity a new competitive axis.

腾讯Hy4Agent开源模型
Back to realtime news

Nearby Updates

All

11/06, 10:02

China unveils an AI agent identity registry at Wuzhen as 20 organisations sign a self-discipline pact

China has unveiled an AI agent identity registry at Wuzhen, with 20 organisations signing a self-discipline pact alongside it, according to Pandaily. The move treats agents as network actors that must be identifiable and traceable, shifting governance from content review to the question of who is acting.

11/06, 09:49

Zetrix AI Launches an Agent Identity Service After RM3.7 Billion in Development Spending

Zetrix AI has launched an AI agent identity service after RM3.7 billion in cumulative development spending, The Edge Malaysia reports. The outlet describes the company as embattled, which makes the launch read as an attempt to convert a large research bet into a commercial product.

11/06, 11:20

Jimeng launches a short-drama studio as Seedance 2.5 falls to 0.15 yuan per second

QbitAI's hands-on review of Jimeng's short-drama studio describes a single workflow covering script writing, storyboard breakdown and visual preview. The piece also notes Seedance 2.5 generation costs as little as 0.15 yuan per second, backed by a long-term partner programme.

11/06, 08:14

Box puts Nvidia's Nemotron at the core of a new agent-first retrieval layer

At BoxWorks on November 5, Box unveiled Box Content Retrieval, which builds Nvidia's Nemotron embedding system into its product so agents can answer queries faster and more accurately. Nvidia expects four-times faster queries, 20% fewer errors and 90% fewer tokens, with the feature rolling out to Enterprise Advance customers early next year.