HONG KONG -- Tencent Holdings announced its latest artificial intelligence large language model, Hy4, on Wednesday, claiming the LLM would "break into the top tier of open source AI models for the first time." The Chinese tech giant said Hy4 outperformed closed-source rivals, such as Anthropic's Claude Opus 5.5, in agentic coding, and surpassed both OpenAI's GPT 6.1 Sol and Claude Opus 5.5 in agentic search and working agent tasks. However, it lagged significantly behind both models and Chinese open-weighted rival Kimi K3.1 Pro on AI-assisted research.

Like most Chinese models, Tencent has pivoted from closed-source to open-source models, drawing in customers to its cloud services. Compared with its predecessor, the Hy4 Preview, the 770-billion-parameter Hy4 LLM can process and analyze images for the first time.

"Although Hy4 is already one of the strongest open-weight models available, there remains a considerable gap between it and leading closed-source models, particularly on our internally developed, high-difficulty benchmarks covering frontier mathematics, long-horizon software development, model iteration and visual question answering," Tencent said.

"We will continue to steadily scale up both pre-training and post-training in an effort to narrow the gap with leading closed-source models." The company did not say when Hy4 would be made available as an open-weight model, where the trained parameters are released as open source, but not the training datasets behind them. But it said it would soon release its Hy4 Flash, an LLM for more everyday use, delivering major boosts to its cost-efficiency and speed in answering queries.

Hy4 is priced between DeepSeek V4 Pro's off-peak and peak rates, with the latter known for its highly competitive pricing.

Cyber security experts have warned that hackers are using AI models to help launch attacks, while Anthropic has been restricting its Claude Mythos models to defensive cybersecurity work, authorized enterprise testing and critical infrastructure protection rather than releasing them to the public.

Tencent said it excluded offensive training content such as vulnerability exploitation, reverse engineering and penetration testing, to avoid strengthening the model's potential for attacks.

In late 2025, when Tencent was widely seen as lagging in the AI arms race, it hired Yao Shunyu, a former OpenAI researcher, as its chief AI scientist, reporting directly to President Martin Lau. Since then, Yao has led development of Tencent's Hy series of large language models and unveiled its first flagship AI model, Hy3, in April.

Tencent's announcement came as the Financial Times reported that Oracle has canceled a five-year lease that would have given the Chinese company access to up to 100,000 advanced AI chips across a range of data centers in Southeast Asia. Tencent did not respond to a request for comment.

Chinese companies have been rapidly releasing cost-effective open-weight AI models to challenge the dominance of U.S. frontier AI labs. Late last month, Moonshot AI released the 2.8 trillion-parameter Kimi K3.1 Pro and its faster variant, Kimi K3.1 Pro Highspeed, which has scored high on agentic coding and frontend coding benchmarks. In August, Alibaba released its latest model, Qwen3.8 Max, and boasted it was "second only to [Anthropic's] Fable 5."

On Artificial Analysis, a well-known independent AI benchmarking platform that recently overhauled its evaluation system, Kimi K3.1 Pro ranks ninth overall, making it the highest-ranked Chinese model in terms of intelligence. Qwen 3.8 Max and DeepSeek V4.1 Flash rank 13th and 14th, or second and third among Chinese models, respectively.