Tencent's Hy4: Open-Source 770B Model Built for Real Work

Tencent has released Hy4 Preview, an open-weight AI model aimed at the work most businesses actually need done: coding, office tasks, financial analysis and research. It is a mixture-of-experts model with 770 billion total parameters, though only 49 billion are active for a given task. That design is intended to provide large-model capability without running the full model every time. Tencent also says it supports more than one million tokens of context, meaning it can handle very large sets of documents, code or internal material in one conversation.

The more important point is where Tencent is putting it. Hy4 is already being used inside its own WorkBuddy and CodeBuddy products, rather than being positioned purely as a benchmark contender. That matters because business buyers should care less about leaderboard scores and more about whether a model can reliably draft reports, analyse financial material, help staff work through documents and support developers. Tencent lists pricing through Tencent Cloud and OpenRouter at $0.834 per million input tokens and $2.501 per million output tokens, which gives operators a clearer starting point for estimating usage costs. You can read the details in Tencent’s announcement.

For owners, this is another sign that powerful models are becoming more available beyond the biggest US providers. The practical question is not whether to switch models this week. It is whether you have clear, governed workflows ready to test across providers. This is the kind of thing we build into an AI command centre, where teams can assess models against real company work, cost and data requirements.