Hy4 preview
Product Hunt Zac Zuo ● Covered by 2 sources
Tencent just previewed Hy4, a 770B open AI model with a huge 1M-token context. It’s built for long-haul agent work, from coding to bug fixing, not quick chat replies.
Based on reporting by Product Hunt, Zac Zuo — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Tencent has put out Hy4 Preview, the latest entry in its Hy model family and the third launch from Tencent Hy. The company is pitching it as a 770B open model for long-horizon work, with a 49B active set and a 1M context window. That is a very specific shape of ambition: less “answer me fast,” more “stay with the job until it’s done.”
The target use cases are the ones that tend to unravel ordinary assistants. Tencent says Hy4 is meant for coding, game development, and complex document analysis. It is also described as an agentic model, which matters because the pitch is not just about generating text or images on demand. It’s about taking a task, working through it, testing along the way, and correcting mistakes before handing anything back.
That framing fits with the broader Tencent Hy family, which the company says covers text, image, video, 3D generation, and understanding. Hy has been built for developers, enterprises, and creators, and the preview reads like another step toward making that stack more useful in actual production work rather than demo clips.
The standout detail here is the autonomy claim. Hy4 is supposed to run its own tests and fix bugs before delivery. If Tencent can make that reliable, it would be more interesting than another model that can talk a good game and then stumble as soon as the task gets long or messy. Long context is nice. Long context that can police itself is the real prize.
My take — AI-written commentary, not fact-checked reporting
Open models keep getting sold as freedom, but the real test is whether they can shoulder boring, stubborn work without theatrics. Tencent is pointing at the right problem here: agents that can stay on task, test themselves, and clean up their own mess. That is less glamorous than benchmark fireworks, which is exactly why it matters.
Read more about this at: Product Hunt