Tencent Hy4 preview is positioned as a next-generation open-source large language model aimed squarely at real-world productivity. Tencent describes it as built for coding, office work, and scientific research, and it was co-designed with internal products like CodeBuddy and WorkBuddy to optimize actual user experience in those scenarios. The model carries 770B total parameters with 49B active parameters and a context window exceeding one million tokens, a configuration that signals a sparse mixture-of-experts style design where a relatively small active subset is routed per query. That combination of broad parameter capacity, modest per-token activation, and an extended context is intended to support long, multi-document workflows while keeping inference tractable.
On independent capability tracking, Hy4 preview lands in the top tier of open models for technical and analytical tasks, ranking seventh in tool calling, eighth in reasoning, tenth in coding, and twelfth in math across large leaderboards. It sits closer to the middle of the pack on long-context and search-style evaluations and trails on legal and finance-specific benchmarks, suggesting it generalizes well to general engineering and research work but is less specialized for highly regulated domains. Distribution is broad: the weights are open-sourced, the model is accessible through Tencent products such as WorkBuddy, CodeBuddy, Yuanbao, and ima, and APIs are offered through Tencent Cloud TokenHub and OpenRouter, giving developers multiple paths to integrate it into productivity tooling.
