Looking at the tencent hy4 preview and trying to work out whether it actually matters? The short answer is just below — I have pulled the launch apart so you do not have to. Tencent dropped this model on 28 August 2026 and open-sourced it on day one, which tells you plenty about where the open-weight race is heading.
The facts in this guide come from TechNode’s report on Tencent Hunyuan’s official launch (published 28 August 2026, fetched 31 August 2026). Where a number is Tencent’s own claim, I say so — clearly.
Short answer:
- Tencent Hy4 preview launched 28 August 2026 as Hunyuan’s next-generation flagship LLM — open-sourced the same day.
- It is a mixture-of-experts model: 770B total parameters, ~49B activated per request.
- Context window exceeds 1 million tokens.
- Free for a limited two-week period inside WorkBuddy and CodeBuddy; also in Yuanbao and ima.
- API access via Tencent Cloud TokenHub and OpenRouter: $0.834/M input, $2.501/M output tokens.
Tencent Hy4 Preview: The Headline Specs
Hy4 preview is Tencent Hunyuan’s new flagship large language model, built for productivity work: coding, office tasks, data analysis, game development and scientific research. The architecture is a mixture-of-experts design — 770 billion parameters in total, with only around 49 billion active for any single request. That is the same efficiency play GLM-5.3 and Kimi K3 are making: frontier-scale capability without frontier-scale inference cost.
| Spec | Tencent Hy4 preview |
|---|---|
| Released | 28 August 2026 |
| Total parameters | 770B (MoE) |
| Active parameters | ~49B |
| Context window | 1M+ tokens |
| Weights | Open — weights + model card on Hugging Face |
| API price | $0.834/M input · $2.501/M output |
How To Try Tencent Hy4 Preview Free
Two genuinely free routes exist right now. First, WorkBuddy and CodeBuddy — Tencent’s productivity and coding assistants — offer free Hy4 preview access for a limited two-week period from launch, so the clock is already ticking. Second, the open weights on Hugging Face mean you can run it yourself if you have the hardware (a 770B MoE is not a laptop job, but quantised community builds tend to follow fast). It is also live inside Yuanbao and ima. For API work, Tencent Cloud TokenHub and OpenRouter both carry it, which makes slotting it into an existing agent stack straightforward.
🔥 Want this set up without the guesswork? Inside the AI Profit Boardroom we’re already building long-context agent workflows on open models like Hy4 — 3,700+ members, four live calls per week, daily tutorials, done-for-you templates and a 30-day roadmap. Or if you’d rather talk it through first, book a free SEO strategy session and I’ll map it to your business personally.
Benchmarks: What Tencent Claims
In Tencent’s internal evaluation, Hy4 preview scored an average of 2.99 out of 4, versus 2.92 for GLM 5.3 and 2.94 for Kimi K3. Tencent also says the model helped optimise parts of its own training and inference systems, lifting end-to-end throughput by 31.8% against a baseline — a genuinely interesting self-improvement claim if it holds up.
⚠️ Vendor-reported numbers: The 2.99/4 evaluation score (vs GLM 5.3 and Kimi K3) and the 31.8% throughput gain are Tencent’s own self-reported numbers from its internal evaluation, per the 28 August launch coverage. No independent benchmarks were available when this was written — treat them as claims, not facts.
For context on how Tencent’s last generation actually performed in the wild, my Tencent HY3 review covers the honest gap between launch claims and daily use — useful calibration before you believe any launch-day chart, and part of why I always recommend a strategy session before betting a workflow on a brand-new model.
Where Hy4 Preview Fits For SEO And Agent Work
The million-token context is the practical story. Whole-site content audits, giant competitor crawls, long agent sessions that do not lose the plot — that is where 1M+ context earns its keep, and until this week your open-weight options for it were basically GLM and Kimi. Hy4 preview adds a third serious player, and the pricing undercuts most Western frontier APIs. If you are already comparing the field, my breakdowns of Qwen 3.8 Flash Next and GLM-5.3-Flash show how quickly this tier is moving — three flagship-class open releases in one month.
Cost is the other half of the argument. At $0.834 per million input tokens and $2.501 per million output tokens, a full-site audit that chews through ten million input tokens costs under nine dollars — the kind of number that changes what you are willing to automate. My rule for any brand-new model still applies here: pick one revenue-relevant workflow, run it side by side against your current model for a week, and let the outputs decide. Launch-day hype is free; switching costs are not.
The Bottom Line On Tencent Hy4 Preview
Tencent Hy4 preview is a serious open-weight flagship: 770B MoE, 1M+ context, day-one open weights and aggressive API pricing. The vendor benchmarks need independent confirmation, and “preview” means things will change — but the two-week free window in WorkBuddy and CodeBuddy makes testing it essentially risk-free. If long-context agent work is on your roadmap, put a real task through it before the free period closes.
If you want help putting open-model agent workflows like Hy4 to work, join the AI Profit Boardroom for the live calls, templates and the 30-day roadmap — or book a free SEO strategy session and we’ll build your plan together.
FAQ: Tencent Hy4 Preview
Is Tencent Hy4 preview open source?
Yes. Tencent released Hy4 preview on 28 August 2026 and open-sourced it the same day, publishing official model weights and a model card on Hugging Face, per TechNode’s launch report.
How big is the Tencent Hy4 preview model?
It is a mixture-of-experts model with 770 billion total parameters, of which about 49 billion are activated for any given request.
What is the context window on Hy4 preview?
Tencent says the context window exceeds 1 million tokens, which puts it in the same long-context class as GLM-5.3 and Kimi K3.
Can I use Tencent Hy4 preview for free?
Yes, for a window: WorkBuddy and CodeBuddy offer free access for a limited two-week period from launch. It is also available in Yuanbao and ima.
What does the Hy4 preview API cost?
API access runs through Tencent Cloud TokenHub and OpenRouter at $0.834 per million input tokens and $2.501 per million output tokens, per the launch coverage.
Is Hy4 preview better than Tencent HY3?
It is the next-generation flagship, a full step up in scale and context length from HY3. Tencent’s own internal evaluation puts it slightly ahead of GLM 5.3 and Kimi K3 — but treat vendor numbers as claims until independent benchmarks land.
Related reading
- Tencent HY3 Review: Honest Test
- Qwen 3.8 Flash Next Benchmarks: The Real Numbers
- GLM-5.3-Flash: Z.AI’s New Vision Model Explained
About the author
Julian Goldie is an SEO agency owner with 10+ years in SEO, 394K+ YouTube subscribers, a 100% Upwork job-success score, 75K+ community members across his groups, and the author of a best-selling SEO book. He publishes daily AI SEO tutorials on YouTube, runs the AI Profit Boardroom community, and offers a free SEO strategy session if you want a custom plan. For agency work, book a call for a custom quote.
Last updated August 2026. This is the living guide to tencent hy4 preview — it gets updated as the tools change.
