Tencent Hunyuan's full next-generation 770B MoE model — released November 4, 2026, now with native visual understanding and Computer Use on top of its coding, analysis, and agentic strengths.
Last verified · 2026-11-05 · by Moe Ameen
Tencent Hy4 is the full version of Tencent Hunyuan's next-generation large language model, released November 4, 2026, following the Hy4 preview that was open-sourced on August 28. It keeps the mixture-of-experts (MoE) architecture the series puts at 770 billion total parameters and adds two capabilities new to the line: native visual understanding and Computer Use. The vision side can read fine image detail, analyze complex charts, and handle visual math; Computer Use lets the model operate software through graphical and command-line interfaces, with specialized training on tools like Blender and CAD that Tencent says improves its ability to plan and carry out long, complex tasks.
Tencent positioned Hy4 as joining the strongest open-source models and said the weights, with a risk-focused System Card, would be released soon — so at launch access was via API through Tencent Cloud TokenHub and OpenRouter (with integrations reported for Hermes and OpenCode) and inside Tencent products like 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; treat those as an early snapshot and confirm them on Tencent Cloud.
On capability, Tencent reported internal gains — an 83.18 average and 96.15% task pass rate in its WorkBuddy office evaluation, a lower financial hallucination rate than the preview, and improved game-development scores — but its comparisons are mainly against other open-source models, not closed frontier systems. Those are vendor results and await independent verification. A faster, cheaper "Hy4 flash" variant is slated to follow.
The key thing for a creator to understand is that Hy4's vision is for understanding, not generating. It reads and reasons over images, charts, and screens, but every output is text or an action — no images, video, audio, captions, or publishing. Its strength is digesting large, technical, or visual material and reasoning over it, which makes it a capable front end. Turning that into finished, on-brand, published content is a separate job.
The full Hy4's fresh edge for creators is that it can finally start from a picture. Point it at the visual raw material a content idea usually hides inside — a webinar's slide deck, a dashboard screenshot, a one-pager, even a rival's best-performing [Carousel](/glossary/hyperframes) — and its native vision reads the thing and returns the actual point in clean text. That is a real unlock, because the hardest part of repackaging visual source material used to be transcribing and interpreting it by hand. But Hy4 stops at that interpretation. It cannot redraw the insight as your own branded visual, put a face on it, or get it in front of an audience. That downstream half is exactly what [Kompozy](/) is built for.
So the pairing runs vision-in, visuals-out. Take the brief Hy4 extracted from a chart or deck and drop it into Kompozy, which generates the net-new visual formats a reasoning model can't: an Infographic Photo or brand-exact Carousel that re-expresses the data as your own graphic, a [Persona Short](/glossary/persona-shorts) where a face-locked HeyGen avatar walks through the finding on camera, and Quote Graphics that lift the single number worth remembering — plus a Blog Article and Email Newsletter for the audience that wants depth. Every piece is rewritten under a [Persona Brief](/glossary/persona-brief) so borrowed source material lands in your voice, not the model's, then [Autopilot](/glossary/autopilot) schedules and publishes the set across the eight social platforms plus blog and email behind a per-post review. Hy4 reads the visual and gets the point; Kompozy turns the point into your own visuals, everywhere.
Hy4 is the full version of Tencent Hunyuan's next-generation large language model, released November 4, 2026, after the Hy4 preview was open-sourced in August. It is a 770B-parameter mixture-of-experts model that adds, for the first time in the line, 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, not creating them — it reads and reasons over images and charts, but it outputs text and actions only, with no image, video, or audio generation. To turn its analysis into visual content you pair it with a generation-and-publishing engine like Kompozy, which makes persona/avatar video, carousels, infographics, quote graphics, blogs, and newsletters.
Tencent positioned Hy4 as an open-source model and said the weights plus a System Card would be released soon; at launch it was accessible via API through Tencent Cloud TokenHub and OpenRouter and inside Tencent products. Reported pricing, unchanged from the preview, is about $0.90 per million input tokens and $2.70 per million output tokens. Confirm the weights timeline and current pricing on Tencent's pages.
The full Hy4 keeps the preview's 770B MoE base and adds native visual understanding and Computer Use, which the preview lacked. The August preview is text-focused and already has public open weights; the full model's weights were promised but not yet released at launch.
Have Hy4 read a chart, deck, or document and return a brief, then bring that text into Kompozy as a source. Kompozy generates 18 formats from that one input — persona/avatar video, carousels, infographics, quote graphics, a blog, and a newsletter — holds a consistent voice and face across them via the Persona Brief, and schedules and publishes across eight social platforms plus blog and email.