Tencent ships Hy4 with vision and computer use, claims second place in China

Tencent has released Hy4, a frontier AI model that understands images for the first time and can operate software on its own — and the company claims it now ranks among the strongest open-weight models in the world, though the weights themselves won't ship until a later, unspecified date. For now, Hy4 is available only through Tencent Cloud, OpenRouter, and Tencent apps such as the Yuanbao chatbot.

The release lands roughly two months after Tencent's preview, and it arrives in a crowded field. Since Tencent shipped Hy3 with 295 billion parameters under the Apache 2.0 license last July, Chinese rivals have pushed hard: Zhipu AI released GLM-5.3, Deepseek put out V4.1-Flash-1017, and Moonshot AI launched Kimi K3.1 Pro. Tencent says it trained Hy4 on significantly more data than its predecessor, and that the model "for the first time" ranks among the strongest open-weight models available.

What can Hy4 actually do?

The headline feature is vision. Hy4 is supposed to recognize fine details in photos and complex diagrams. In one demo, the model finds a roughly two-meter component on a warehouse roof inside a massive aerial image. That's a needle-in-a-haystack task that older models routinely fumbled.

The second big addition is computer use. Hy4 can control software on its own, and Tencent trained it specifically on professional tools like Blender and CAD programs, plus tasks that require chaining multiple applications in sequence. That puts it in direct competition with agentic offerings from OpenAI and Anthropic, which have made computer control a central selling point of their latest models.

Image understanding also feeds into coding. Tencent says Hy4 can build websites and games, and for game development the company worked with its own gaming studios to adapt the model for Unreal, Unity, and Godot. On the mobile game Honor of Kings, the dev team reports Hy4 handles configuration tasks nearly flawlessly and finds the root causes of bugs far more often than before.

Is finance the real proving ground?

Tencent put particular effort into finance and office work, and it backed the claim with an unusual test: more than 200 finance professionals evaluated Hy4 on nearly 600 real-world tasks. According to Tencent, Hy4 ranked ahead of Chinese rivals Deepseek V4.1 Flash, Kimi K3.1 Pro, and GLM-5.3 Flash.

The company also claims a sharp drop in hallucinations during web research. A revamped search feature cut the hallucination rate from nearly 16 percent to about 5 percent in internal tests. Hy4 can collaborate with humans in real time on Word, Excel, and PowerPoint files.

Partner companies report similar progress, though they lean on their own internal benchmarks. Financial platform Comein sees Hy4 as the only Chinese model on par with Anthropic's Claude Opus 5.5 in second place, and clearly ahead of OpenAI's GPT-6.1 Sol on data queries. Agent platform Manus reports that Hy4 completes nearly all document tasks successfully.

What do independent tests say?

The first independent assessment, from benchmarking firm Artificial Analysis, paints a more sobering picture. On its Intelligence Index — which combines knowledge work, coding, science, and general knowledge — Hy4 scores 35 points, above the median of 30. That ties it with Grok 4.7 and places it second among Chinese lab models, behind Kimi K3.1 Pro at 46 points. In Tencent's own finance test, Hy4 still ranked ahead of Kimi.

The numbers get worse without web search. On the AA-Omniscience factual accuracy test, Hy4 answers about half the questions correctly from its training data. When it doesn't know an answer, it makes one up 89 percent of the time instead of admitting the gap. For buyers weighing Hy4 against Western competitors, that figure matters more than any finance leaderboard.

The broader gap is widening, too. Artificial Analysis revamped its index in late October, with 60 percent of it now based on private test sets. Since that change, the distance between open models and the top of the leaderboard has grown to 12 points, up from 3.

Can AI help build the next version of itself?

Tencent is also positioning Hy4 as an AI research accelerant. In an internal test spanning research tasks across seven fields, the company says Hy4 ranks first among openly available models. Together with its agent system Hyra 2.0, Hy4 optimized Tencent's own training pipelines for smaller models.

It then went a step further: Hy4 independently trained its own version of Jev, a decision-making model from US startup TypeSafe AI, and Tencent says that version outperforms the original on several unnamed tests. The company is releasing two specialized models for biology and physics, created this way, as open models. Going forward, Hy4 is meant to support and speed up AI development inside Tencent itself.

What about safety and pricing?

With the weight release planned, Tencent is stressing safety. The company says it deliberately filtered out hacking-related material during training, and its own security labs found no elevated risk of misuse. The model also falls well below Tencent's risk threshold for biological hazards, according to the company. A detailed safety report will follow alongside the weights.

Pricing stays flat despite the improvements. One million input tokens costs about $0.83, and one million output tokens runs about $2.50 — the same as the preview. Tencent admits Hy4 still falls well behind the best commercial models on hard tasks, including advanced math, longer coding projects, and complex image analysis. A faster, cheaper variant called Hy4 Flash is planned next.

Why does the weight release timing matter?

The timing of the open-weight release collides with tightening regulation. China's commerce and science ministries published a draft export control framework in October that would cover model weights for any model ranked in the global top 10 or exceeding two trillion parameters. The draft does include a fast-track process for open-source releases, with automatic approval — but until the rules are final, the fate of Hy4's weights remains an open question, and Tencent hasn't named a release date.