Tencent Hy4: what is known about the open-weight model, its capabilities and API access

Published: November 4, 2026 · Official

On 4 November 2026 the South China Morning Post reported that Tencent had launched Hy4, an open-weight model the company bills as top-tier. Tencent has confirmed a coming general release of Hy4 and is preparing the cheaper Hy4 Flash. Licence terms and exact specifications have not been disclosed.

What Tencent actually announced

The SCMP piece ran on 4 November 2026 and frames Hy4 as Tencent's bid to break into the top tier of open-weight models. Days earlier, at the 2026 Global Digital Ecosystem Summit in Shenzhen, the company confirmed that the general release of Hy4 is coming soon and announced it was preparing Hy4 Flash.

The positioning was put by Fred Sun, Paris-based general manager of Tencent Cloud Europe: he said there would be interesting news around the new model, which would put the company more or less, and definitely, in tier one of open-weight models. In 2025 Tencent was seen as a latecomer to large language models; Tencent Cloud now argues that is no longer the case.

Capabilities: what is officially claimed

Tencent describes Hy4 as built for practical, real-world work rather than synthetic exercises. The official release lists specific areas of improvement over the previous Hy generation.

• Higher reasoning efficiency.
• Stronger performance on long-horizon tasks that must run to completion across many steps.
• Better results on real-world productivity tasks.
• Visual understanding, a first for the Hy line.
• Hy4 Flash, a cost-effective model that carries over Hy4's visual understanding and agentic capabilities.

What the sources do not say: licence and specs

Hy4 is described as an open-weight model, meaning the weights should be available to download and self-host, unlike models that exist only behind an API. Openness and choice are exactly what Tencent Cloud Europe is pitching to European customers, alongside on-premises options and sovereign cloud offerings.

The licence terms, however, are not disclosed in the available publications: no licence type, no commercial-use restrictions, no attribution requirements. Nor are the architecture, parameter count, context size or benchmark results given. Tencent promises stronger performance but offers no measurements to check it against.

Until the licence is published, it is too early to judge how freely Hy4 can be embedded in products. That is the key open question alongside the weights themselves.

API access: Tencent Cloud TokenHub

The stated route to the model is Tencent Cloud TokenHub, a unified platform with a single API. According to the company, it hosts both its own Hy models and third-party ones; in an interview with diginomica, Fred Sun names DeepSeek, GLM, Kimi and Minimax. Access management, usage, governance and billing are all handled in one place.

Tencent stresses that it releases the latest open-weight models through its public cloud API among the first. For customers the argument is right-sizing: rather than sending everything to expensive frontier models, pick the best fit for coding, inference, image generation or agentic workloads.

The scale Tencent cites: Tencent Cloud serves customers in more than 150 countries and regions, its overseas customer base has doubled over the past year, and international demand for its AI products and services has grown by triple digits. The WorkBuddy agent has handled over nine billion task requests since launching in March, spanning more than 50 industries and 50% of China's top 500 enterprises, with an ecosystem of over 50,000 developers and partners and more than 200,000 skills.

What to use right now

While Hy4 and Hy4 Flash work their way to broad availability, similar jobs, from code and long documents to agentic workflows, are already covered by models in the API Market catalogue: one key for every model, payment in roubles, no VPN.

• deepseek-v4.1-flash: code and text at a low price.
• glm-5.3: code and technical writing.
• kimi-k3.1-pro: code, agents and long context.

What comes next

On paper Tencent now has everything needed to count as a tier-one player in open weights: its own models, a single API gateway that also carries rivals' models, and cloud presence in more than 150 countries. What outside observers still lack is the substance: the licence, the specifications and independent measurements.

The next step is the general release of Hy4 and the arrival of Hy4 Flash. Tencent gives no dates, saying only that it is coming soon. Keep an eye on the API Market model catalogue, which is updated regularly and already carries comparable models.

Models you can use right now

Connect in a minute

from openai import OpenAI

client = OpenAI(
    base_url="https://api-market1.shop/v1",
    api_key="sk-your-key",
)

reply = client.chat.completions.create(
    model="deepseek-v4.1-flash",
    messages=[{"role": "user", "content": "Hi! Tell me about yourself."}],
)
print(reply.choices[0].message.content)

FAQ

Sources

• Tencent launches Hy4 in bid to break into AI’s top tier of open-weight models - South China Morning Post — South China Morning Post, 2026-11-04
• Tencent Cloud Accelerates International Growth, Advances Enterprise AI and Expands Global Partnerships — finance.yahoo.com, 2026-10-29
• 2026 Global Digital Ecosystem Summit – late to the AI party, but first on the European dancefloor? — diginomica.com, 2026-10-30

More news

• Gemini 4 Hydrogen: what Google launched on November 12
• Gemini Decision 1: Google ships a decision model at $0.10 per 1M tokens
• Gemini 3.7 Flash: Google bets on super-fast coding and half the price
• Gemini 4 Pro leaks promise big benchmarks and a low price
• Qwen 4: confirmed facts, the latest teaser and the release status of Alibaba's new model

Русская версия