Tencent Hunyuan's open-source, next-generation MoE model — 770B total parameters, ~49B active, and a 1M-token context — released August 28, 2026 and built for coding, agentic engineering, analysis, and research.
Last verified · 2026-08-29 · by Moe Ameen
Tencent Hy4 preview is a large language model from Tencent's Hunyuan team, released and open-sourced on August 28, 2026. It is a mixture-of-experts (MoE) design with 770 billion total parameters and roughly 49 billion active per token, paired with a context window Tencent describes as exceeding 1 million tokens. It is aimed at productivity and technical work — software engineering (debugging, long-context code tasks), office and financial analysis, game development, and scientific research such as AI R&D and physics simulation — rather than at chat or creative content.
Tencent published the model weights openly on Hugging Face, ModelScope, GitCode, and CNB, alongside a lower-precision Hy4 preview-FP8 variant. It also ships inside Tencent products including WorkBuddy, CodeBuddy, Yuanbao, and ima, and is available via API through Tencent Cloud TokenHub and OpenRouter, with pricing listed around $0.834 per million input tokens and $2.501 per million output tokens and a limited free window at launch. Treat those numbers as an early snapshot and confirm them on Tencent Cloud.
On capability, Tencent reports gains across a set of engineering benchmarks and an internal blind evaluation averaging 2.99 out of 4.00 across 203 tasks judged by 163 experts — figures that are the vendor's own and await independent verification. The most distinctive claim is agentic: Tencent says Hy4 participated in optimizing its own training methods, data strategies, and evaluation, and autonomously tuned its inference for a reported 31.8% throughput improvement.
The honest framing for a creator: Hy4 is a text-in, text-out reasoning engine, not a content studio. It produces text and code — no images, video, audio, captions, or publishing. Its strength is digesting large or technical inputs and drafting from them, which makes it a capable front end. Turning that text into finished, on-brand, published content is a separate job.
Hy4's sharpest edge for creators is translation: it can take something dense and authoritative — a technical whitepaper, a compliance doc, a quarter of financials — and reason across the whole thing in one pass to produce an accurate, plain-language draft. That is a hard, valuable step, and it is where expert-led content usually stalls. But Hy4 stops at correct text on a screen. It cannot put a face on that explanation, break it into a scroll-stopping visual sequence, or get it in front of anyone. That downstream half is exactly what [Kompozy](/) is built for, and the pairing plays to each tool's strength.
Take the explainer Hy4 drafted from your technical source and drop it into Kompozy. Because the hardest content to make watchable is the dry, expert kind, Kompozy's most useful moves here are the ones Hy4 can't touch: a [Persona Short](/glossary/persona-shorts) narrates the concept on camera with a face-locked HeyGen avatar, a brand-exact [Carousel](/glossary/hyperframes) walks the complex point through step by step, and Quote Graphics lift the one line worth remembering — while a matching Blog Article and Email Newsletter go long for the audience that wants depth. Every piece is rewritten under a [Persona Brief](/glossary/persona-brief) so authoritative 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. For privacy-sensitive teams, Hy4's open weights mean the drafting can happen on your own infrastructure before Kompozy handles the production and distribution. Hy4 makes the hard idea correct and clear; Kompozy makes it watchable, on-brand, and everywhere.
Hy4 preview is Tencent Hunyuan's next-generation large language model, released and open-sourced on August 28, 2026. It uses a mixture-of-experts architecture with 770 billion total parameters (about 49 billion active) and a context window exceeding 1 million tokens, built for coding, agentic engineering, office and financial analysis, game development, and scientific research.
No. Hy4 is a text-in, text-out model — it outputs text and code only, with no images, video, or audio. To turn its scripts and briefs into visual content you pair it with a generation-and-publishing engine like Kompozy, which makes persona/avatar video, carousels, images, quote graphics, blogs, and newsletters.
Tencent open-sourced the model weights (plus an FP8 variant) on Hugging Face, ModelScope, GitCode, and CNB. API access was listed around $0.834 per million input tokens and $2.501 per million output tokens via Tencent Cloud TokenHub and OpenRouter, with a limited free window at launch. Confirm the license and current pricing on Tencent's pages.
It is tuned for engineering and analysis rather than English short-form voice, so it works best as an upstream drafting and research brain — especially for turning technical or long source material into a clear draft. For on-brand captions, formatting, and distribution, hand that draft to a content engine like Kompozy.
Draft or summarize in Hy4, then bring the text into Kompozy as a source. Kompozy generates 18 formats from that one input — persona/avatar video, carousels, quote graphics, photo posts, 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.