Observed Signal · Jul 10, 2026 · Technical Release · Source: Hello China Tech · Impact: 3/5 · Sentiment: Neutral
Tencent Releases Hy3; WeChat Agent Uses Different Model
Tencent announced Hy3 on July 6, 2026 — its first full public model release under Chief AI Scientist Yao Shunyu. Hy3 is a 295-billion-parameter model that activates about 21 billion parameters per call; Tencent reports internal WorkBuddy task success rising from 72% to 90% and hallucination rates falling from 12.5% to 5.4%. API pricing was published at RMB 1 per million input tokens and RMB 4 per million output tokens. Brokers such as Bernstein view Hy3 as sufficient to enable agent-driven commerce inside WeChat (reported at 1.43 billion MAUs), but reporting from outlets including Caixin, 36Kr, GeekPark and Bloomberg indicates WeChat’s limited-test agent (Xiaowei) currently uses WeLM (and selected DeepSeek services) rather than Hy3, exposing an organizational integration gap that could delay commercialisation inside Tencent’s main consumer surface.
Hy3 is a new foundational model from a major Chinese tech company and could enable agent-driven commerce inside WeChat — a large ad/commerce surface — but the reported lack of Hy3 integration with WeChat’s agent (Xiaowei) creates uncertainty about timing and commercial impact.
Track Tencent Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Tencent publicly released Hy3 on 2026-07-06 as its first full model release under Chief AI Scientist Yao Shunyu.
- Hy3 has 295 billion parameters and activates ~21 billion parameters per call (deliberate mid-range design).
- Internal tests reported task success rates rising from 72% to 90% and hallucination rates dropping from 12.5% to 5.4%.
- API pricing published at RMB 1 per million input tokens and RMB 4 per million output tokens.
- Multiple outlets report WeChat’s agent Xiaowei is currently powered primarily by WeLM (and selected DeepSeek tasks), not Hy3.
Connected Companies & Entities
4 Entities mapped“On July 6, Tencent released Hy3, the first full model release under Chief AI Scientist Yao Shunyu....”
“Bernstein maintained its Outperform rating and HK$780 price target after the release....”
“Multiple outlets, including Caixin, 36Kr, GeekPark, and Bloomberg-sourced English reporting, identify WeLM... as the primary engine....”
“Several also report DeepSeek handling selected tasks....”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Tencent's Agent Flood Tests Its AI Catch-Up Strategy
Between March and May 2026, Tencent rolled out a broad wave of AI agent products across office productivity, coding, desktop control, data analytics, design, and OS-level assistance — including WorkBuddy, CodeBuddy, QClaw, DataBuddy, Ardot and Marvis. Marvis, released in late May, implements a six-agent architecture with a main agent delegating to File, Computer, App, Browser and Search agents. Tencent also added agent capabilities to Yuanbao, QQ Browser, Sogou Input and Tencent Meeting and is preparing to convert WeChat Mini Programs into AI-callable skills. Company leaders (Pony Ma and Tang Daosheng) framed the shift from chatbots to task-oriented agents. The product push arrives amid mixed metrics: Yuanbao reported 57.35M MAUs versus ByteDance’s Doubao at 345M and Alibaba’s Qwen at 166M; Tencent Cloud ranked 5th with ~8% IaaS share in China. Tencent estimates the new AI products imply roughly RMB 8.8 billion ($1.2 billion) quarterly non‑IFRS operating-profit drag.
Tencent Rapidly Deploys OpenClaw Agents on WeChat
Tencent launched three OpenClaw-based AI agent products on March 9—QClaw (desktop app linking OpenClaw to WeChat and QQ), WorkBuddy (an enterprise agent for WeChat Work, Feishu and DingTalk), and an OpenClaw integration for WeChat Work’s smart bot platform. The launches prompted a near 7% jump in Tencent’s stock the following Monday. WorkBuddy experienced outages on release, and Tencent Cloud scaled servers tenfold and apologized. The article positions Tencent’s move as urgent and strategic: after heavy infrastructure spending and talent hires, Tencent aims to compete via agent integrations in WeChat rather than by building a superior foundation model, leveraging messaging‑layer control to offset gaps in model rankings and consumer app metrics.
Tencent Tests Xiaowei AI Assistant in WeChat
Tencent said it is testing Xiaowei, a native AI assistant, on a small scale inside Weixin (the Chinese version of WeChat). The assistant can be interacted with via text or voice, helps users communicate with friends and can launch WeChat mini-programs. Tencent’s move is part of a broader push to expand AI capabilities and monetization inside its massive WeChat/Weixin ecosystem, which the company says has more than 1.4 billion monthly active users combined. The test aims to help Tencent compete with other Chinese AI players such as Alibaba, DeepSeek and Zhipu. Tencent has developed its own model family under the Hunyuan brand and recently hired an OpenAI researcher to serve as its chief AI scientist, though the company did not disclose technical details of Xiaowei or the underlying models.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.