+1 (415) 347-6981Get started
Fixed prices · You own the code · Proposal in 24h hello@startrise.io →
AI Model Release Tracker China · updated
  • 13Releases tracked
  • Nov 16Latest · Hy4 flash
  • 69%Open-weight
  • CNShenzhen, China

Tencent Hunyuan, now known as Tencent Hy, is the AI model family of Tencent, the Shenzhen, China-based internet and technology company founded in 1998. Its flagship Hy3, launched in July 2026, is a hybrid fast-and-slow-thinking mixture-of-experts model with 295 billion total and 21 billion active parameters, released under the Apache 2.0 license. Hy3 powers Tencent products including Yuanbao, CodeBuddy and ima. The family also includes the HunyuanImage 3.0 image model, the Hy4 preview, and specialist open models such as Hy-Memory-30B-A3B and Youtu-Parsing-Omni. Most tracked releases are open-weight. The earlier Hunyuan-T1 reasoning model was closed.

Tencent Hunyuan

China hunyuan.tencent.com

13 releases latest Nov 16, 2026

  1. Hy4 flash

    Cheapest Hy model yet: 224B total (64B Engram) with 12B active, multimodal input, over 150 tps decode. Live on Tencent Cloud TokenHub, WorkBuddy and Yuanbao; Tencent plans to open-source it.

    • Multimodal
    • API
    • App
    • 1M ctx
  2. Hy Realtime

    Real-time voice model in the Hunyuan line with agentic tool calling (web search) and singing. Live in Yuanbao voice calls and Tencent customer service; Tencent Cloud API coming soon.

    • Audio
    • App
  3. Hy4

    Adds vision and computer use (CLI and GUI) to the Hy line; about $0.90/$2.70 per 1M tokens. Tencent says weights and system cards follow in the coming days.

    • Multimodal
    • API
    • App
  4. Hy-SimulMT

    30B-A3B real-time simultaneous speech translation model for Chinese and English in both directions, with real-time voice cloning; FP8 and GGUF builds, also in the Hy Translate app.

    • Audio
    • Open weights
    • App
  5. Hy-Nomas

    Apache-2.0 interactive video world model: generates a world segment by segment, steered by multi-turn text instructions and camera control. Ships with Hy-Nomas-Bench.

    • Video
    • Open weights
  6. WeVisOne-VL (2B and 4B)

    WeChat Vision model built on Qwen3-VL that unifies detection, referring localization, counting, pointing and OCR with referring segmentation and alpha matting. Apache-2.0 code.

    • Multimodal
    • Open weights
  7. Hy-Memory-30B-A3B

    MoE long-term-memory writer (~3B active) that replaces the general LLM inside Mem0, MemOS and Hy-Memory; Apache 2.0.

    • LLM
    • Open weights
  8. Youtu-Parsing-Omni

    5B omni-modal parsing model: documents, images, charts, geometry, audio and video → structured JSON; SOTA OmniDocBench v1.6 (96.96).

    • Multimodal
    • Open weights
  9. Hy4 preview

    • Reasoning
    • Open weights
  10. Hy3

    295B MoE, Apache 2.0.

    • Reasoning
    • Open weights
    • API
  11. Hy3 preview

    • Reasoning
    • Open weights
  12. HunyuanImage 3.0

    • Image
    • Open weights
  13. Hunyuan-T1

    • Reasoning
    • API
    • App

Tencent Hunyuan FAQ

What is Tencent Hunyuan's latest AI model?
Tencent Hunyuan's most recent release in the Startrise tracker is Hy4 flash, released Nov 16, 2026.
How many AI models has Tencent Hunyuan released?
Startrise tracks 13 notable Tencent Hunyuan releases since January 2025, from Hunyuan-T1 (Mar 21, 2025) to Hy4 flash.
Does Tencent Hunyuan release open-weight models?
Yes. 9 of 13 tracked Tencent Hunyuan releases (69%) ship with downloadable weights; the most recent is Hy-SimulMT.
Where is Tencent Hunyuan based?
Tencent Hunyuan is based in Shenzhen, China.