On November 4, 2026, Tencent Hunyuan shipped the full version of Hy4 — its 770B mixture-of-experts model — adding, for the first time, native visual understanding and the ability to operate a computer through graphical and command-line interfaces.
2026-11-05 · by Moe Ameen
On November 4, 2026, Tencent's Hunyuan team officially released Hy4, the full version of its next-generation large language model. It follows the Hy4 preview Tencent open-sourced on August 28, 2026, and keeps the same headline shape — a mixture-of-experts (MoE) design that the Hy4 series puts at 770 billion total parameters. The release's two new capabilities are what separate it from the preview: native visual understanding and Computer Use, both appearing in the Hy4 line for the first time.
On vision, Tencent says Hy4 can recognize fine image detail, analyze complex charts, and handle visual math — reading a diagram or screenshot and reasoning over it rather than working from text alone. On Computer Use, Tencent describes the model executing tasks through both graphical interfaces and the command line, with specialized training on tools like Blender and CAD that Tencent says improves its ability to plan and carry out long, complex tasks. In other words, Hy4 is positioned as a model that can both see a screen and act on it.
Tencent framed Hy4 as joining the ranks of the strongest open-source models and said the model weights, along with a System Card addressing risks such as cybersecurity and biosecurity, will be released soon; Tencent says it trained on defensive security data — vulnerability remediation, reverse engineering, and penetration testing for analysis purposes — while explicitly excluding offensive exploit-development material, and its internal Xuanwu and Keen Security labs reported no significant concerns in the areas they evaluated. At launch the model is available through Tencent Cloud TokenHub and OpenRouter (with integrations reported for Hermes and OpenCode) and inside Tencent products including WorkBuddy, CodeBuddy, Yuanbao, and ima. Tencent says API pricing is unchanged from the preview at about $0.90 per million input tokens and $2.70 per million output tokens, with cached input at about $0.045 per million tokens; Tencent Cloud TokenHub billed at 50% through December 31, 2026, with a free window in WorkBuddy and CodeBuddy from November 4–15. A faster, cheaper "Hy4 flash" variant is slated to follow.
Tencent also reported internal evaluation gains — an 83.18 average and a 96.15% task pass rate in its WorkBuddy office test, a higher score than several named peers in a crowdsourced financial evaluation, and a drop in financial hallucination rate from 12.95% in the preview to 7.2% — though its comparisons are mainly against other open-source models, not closed frontier systems. Those figures are Tencent's own; treat them as vendor claims until independent results land, and confirm specs, pricing, and the weights timeline on Tencent's official pages.
There are two ways to act on this today. The fast one rides the story: paste "Tencent just released the full Hy4 with native vision and Computer Use" into [Kompozy](/)'s Quick Ingest and fan that single beat into a Text Post, an X thread, a LinkedIn [Carousel](/glossary/hyperframes) explaining what a see-and-act model changes for builders, a captioned [Persona Short](/glossary/persona-shorts), and a blog explainer — then schedule and publish the set across the eight social platforms plus blog and email. Kompozy is a generation-and-publishing engine, so one launch becomes a full cross-platform package the same afternoon.
The durable read is simpler: as models learn to see and act, the scarce part of a content operation isn't the reasoning, it's the finished, on-brand output and the distribution. Hand Hy4 a dense or visual source and let it return a clean brief, then drop that into Kompozy, which does the half a reasoning model can't — it generates across 18 formats, rewrites every one under a [Persona Brief](/glossary/persona-brief) so it sounds like you rather than a raw model, puts a face on the idea with a HeyGen avatar, and ships the batch through [Autopilot](/glossary/autopilot) behind a per-post review. If you already use the open-sourced [Hy4 preview](/ai-tools/tencent-hy4-preview) for drafting, the full model is a stronger front end for the same workflow.
Hy4 is the full version of Tencent Hunyuan's next-generation 770B mixture-of-experts model, released November 4, 2026, after the Hy4 preview was open-sourced on August 28. The headline additions over the preview are native visual understanding — reading charts, image detail, and visual math — and Computer Use, operating software through graphical and command-line interfaces.
No. Hy4's vision is for understanding inputs — it can read and reason over images, charts, and screens, but it still outputs text and actions, not images, video, or audio. To turn its analysis into visual, multi-format content you pair it with a generation-and-publishing engine like Kompozy.
Tencent positioned Hy4 as an open-source model and said the weights, with a System Card covering cybersecurity and biosecurity risks, will be released soon. At launch it is accessible via API through Tencent Cloud TokenHub and OpenRouter, and inside Tencent products like WorkBuddy, CodeBuddy, Yuanbao, and ima. Confirm the weights timeline and license on Tencent's pages.
Tencent says API pricing is unchanged from the preview: about $0.90 per million input tokens and $2.70 per million output tokens, with cached input at about $0.045 per million tokens. Tencent Cloud TokenHub was billing at 50% through December 31, 2026. Verify current figures on Tencent Cloud before budgeting.