Tencent Hunyuan
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | Turbo S: ¥0.8/M tokens, Turbo: ¥1.2/M, Lite: ¥0.3/M | per million tokens |
| Output | Turbo S: ¥0.8/M tokens, Turbo: ¥4.8/M, Lite: ¥0.3/M | per million tokens |
🤖 Supported Models (9)
✨ Pros
- ✓Hunyuan Hy4 preview: open-weight 770B/49B MoE, native 1M context, Apache 2.0 license
- ✓Hunyuan Turbo S ¥0.8/M token, great value
- ✓Native Function Calling via Hunyuan Functions
- ✓Hunyuan Role for character-driven conversations
- ✓Rich Tencent ecosystem integration (WeChat/Tencent Cloud)
- ✓Direct China access, backed by Tencent Cloud
- ✓Image understanding + generation (Hunyuan Vision)
⚠️ Cons
- ×Model quality behind Alibaba Qwen3.5-Max and DeepSeek-V3
- ×Weaker English capabilities
- ×API documentation messy (Tencent Cloud style)
- ×Some models enterprise-only
- ×Higher token consumption than comparable models
🎯 Best For
Tencent Cloud ecosystem customers; Function Calling scenarios; role-playing/customer service; cost-priority Chinese apps
💰 Pricing & Plans
| Model | Input (¥/M tokens) | Output (¥/M tokens) | Best For |
|---|---|---|---|
| Hunyuan-a13b | ¥0.5 | ¥2 | Open-weight reasoning (A1 family), cheap deep thinking |
| Hunyuan-role-latest | ¥2.4 | ¥9.6 | Role-play and character-driven chat |
| Hunyuan-translation | ¥1.2 | ¥3.6 | High-quality translation |
| Hunyuan-translation-lite | ¥1 | ¥3 | Low-cost translation |
| Tencent HY Vision 1.5 Instruct | ¥3 | ¥9 | Image understanding |
| Hunyuan-turbos-vision | ¥3 | ¥9 | Fast vision understanding |
| Hunyuan-t1-vision | ¥3 | ¥9 | Vision plus turbo reasoning |
| Hunyuan-turbos-vision-video | ¥3 | ¥9 | Video understanding |
| Hunyuan-embedding | ¥0.7 | ¥0.7 | Embeddings for RAG |
| Hunyuan Turbo S (legacy) | ¥0.8 | ¥0.8 | Low-latency chat; still billed for existing services |
🔧 API & Developer Experience
- •API Style: Tencent Cloud-native REST endpoints with TC3-HMAC-SHA256 request signing, plus an OpenAI-compatible chat mode (混元兼容接口) so most OpenAI SDKs work after a base-URL and key swap.
- •SDKs & Tooling: First-party Tencent Cloud SDKs for Python, Java, Node.js, Go, PHP and .NET, an interactive API Explorer, and OpenAI-compatible clients for the wider ecosystem.
- •Tool Calling & Agents: Native function calling via Hunyuan-Functions, role-play through Hunyuan-role-latest, and SSE streaming for agent-style streaming UIs.
- •Multimodal Breadth: Vision (HY Vision 1.5, turbos-vision, t1-vision), video understanding (turbos-vision-video), translation models and embeddings all available from one console.
- •Console & IAM: Tencent Cloud Console with CAM sub-accounts and API keys; activation requires personal or enterprise real-name verification.
- •Billing & Quotas: Postpaid daily settlement or prepaid resource packs; per-account QPS quotas; a free 1M-token trial pack on first activation.
- •TokenHub Migration: Tencent is moving Hunyuan services to the TokenHub platform — new model capabilities and new purchases go through TokenHub, while existing services keep running on the legacy console.
🏢 Tencent Ecosystem & Open-Weight Models
Tencent's edge is distribution: Hunyuan models sit inside the company that runs WeChat, QQ, Yuanbao, WeCom, Tencent Meeting and China's largest gaming and ad platforms, so a Hunyuan API call plugs directly into the ecosystem your users already live in — WeChat mini-programs, WeCom bots, or Tencent Cloud-hosted backends. On the model side, Tencent pairs closed API models with genuinely open weights: the Hunyuan A1 reasoning family ships under permissive licensing and is a popular self-hosted alternative to DeepSeek for RAG and agent pipelines. Combined with a broad multimodal lineup — vision, video understanding, image/video generation and 3D — Hunyuan is less a single model than a Tencent-grade platform play.
🌐 China Access & Latency
Tencent Hunyuan is a first-party Tencent Cloud product, which makes it one of the lowest-friction domestic APIs for mainland teams. Endpoints run in Tencent Cloud data centers across mainland China, so calls from Chinese production environments stay inside the country — no proxy, no cross-border egress, and typical RTT in the low single-digit to low-teens milliseconds from major mainland cities. Billing is in CNY with postpaid daily settlement, and activation requires Tencent Cloud real-name verification (personal or enterprise), which also unlocks the free 1M-token trial pack. The WeChat deep-link is the real differentiator: products needing WeChat login, payment or messaging can keep the whole stack on Tencent Cloud. Caveats: international developers face real-name signup and Chinese-first documentation, and new model services are migrating to the TokenHub platform.