Overview
Hunyuan is Tencent’s home-grown large language model family, launched in 2023 and now spanning text, vision, speech, and 3D. For most users the entry point is the free consumer chat at hunyuan.tencent.com, while developers reach the same models through Tencent Cloud’s API, which is in the process of migrating to the unified TokenHub console. What sets Hunyuan apart in the crowded Chinese-model field is breadth: instead of one flagship, Tencent ships a ladder of specialized models — Hunyuan-T1 for deep reasoning, TurboS for low-latency responses, the a13b lightweight general model, dedicated translation and role-play variants, a vision-instruct model, and an embedding model for search. In our testing the Chinese fluency and logical reasoning sit comfortably in the top tier of domestic models, and the free Hunyuan-lite tier makes it the easiest Chinese LLM to try at zero cost. The main friction is operational: Tencent has been consolidating billing into TokenHub, so prices are scattered across pages and occasionally contradictory, and cloud API use requires real-name verification.
Key Features
- Tiered model menu — Pick T1 for step-by-step reasoning, TurboS for speed, or a13b for cheap general tasks, each tuned for a different job rather than one compromise model.
- Free Hunyuan-lite + 1M-token starter pack — The lite model is permanently free and new cloud accounts get a shared 1M-token trial valid for a year, so evaluation costs nothing.
- Multimodal and vertical variants — Vision-instruct, translation (in and out of Chinese), role-play, and embedding models let you stay inside one vendor for RAG and agent pipelines.
- Open weights available — Several Hunyuan models are released with open weights, enabling self-hosting and fine-tuning for teams that need data control.
- Deep Tencent ecosystem ties — Native fit with WeChat mini-programs, Tencent Cloud, and enterprise WeChat for Chinese business workflows.
Pricing
| Plan | Price | For |
|---|---|---|
| Free (Hunyuan-lite) | ¥0 | Individuals, light chat and drafting |
| Starter pack | 1M tokens free | New cloud accounts, 1-year validity |
| Pay-as-you-go (a13b) | ¥0.5 in / ¥2 out per 1M tokens | Production text at low cost |
| Pay-as-you-go (TurboS) | ¥0.8 in / ¥2 out per 1M tokens | Latency-sensitive apps |
| Pay-as-you-go (T1) | ¥1 in / ¥4 out per 1M tokens | Deep reasoning tasks |
Comparison
Compared to DeepSeek, Hunyuan trades some open-source purity for a wider first-party model menu and tighter Tencent product integration; DeepSeek remains the leaner choice for self-hosting. Against Qwen, both are strong Chinese models, but Qwen’s open-weight ecosystem is larger, while Hunyuan’s free lite tier is more generous. Versus Kimi, Hunyuan wins on model variety and translation, whereas Kimi’s long-context handling is the cleaner pick for massive-document work.