Research

Tencent Hunyuan Model Lineage 2026

Tencent Hunyuan release history 2024-2026: Hunyuan-Large (Nov 2024), Hunyuan-Video (Dec 2024), HunyuanWorld 1.0 + 1.5, Tencent HY 2.0 (Dec 2025), Hy3preview (Apr 2026), Hunyuan 3.0.

By Ramanath, CTO & Co-Founder at Presenc AI · Last updated: May 2026

What this is

Tencent's Hunyuan family is the most modality-diverse Chinese AI lineage: text, image, video, 3D, and immersive world models. Open-source across most categories since 2024. This page is a 2026-05-15 release-by-release reference.

Hunyuan Release Timeline

DateReleaseNotable
May 14, 2024Hunyuan text-to-image (DiT)Open-source, free commercial use
2024 (mid)Hunyuan MoE architectureIntegrated into Tencent collaboration products
Nov 5, 2024Hunyuan-Large + Hunyuan3D-1.0MoE LLM + 3D generation, open-source
Dec 3, 2024Hunyuan-Video13B params; open-source text-to-video
Jul 26, 2025HunyuanWorld-1.0First open-source immersive 3D world generation
Dec 8, 2025Tencent HY 2.0 (Hunyuan 2.0)406B params (32B active), 256K context
Dec 18, 2025HunyuanWorld-1.5 (WorldPlay)Real-time 3D world creation
Apr 16, 2026HY-World-2.0State-of-the-art 3D world model
Apr 23, 2026Hy3preview (LLM)295B params (21B active), 256K context, fast+slow thinking
Apr 2026Hunyuan 3.0 launchSignificant programming capability improvement; WeChat AI agent

Hunyuan Family at a Glance (Mid-2026)

FamilyCategoryOpen / ProprietaryBest at
Hunyuan / Hunyuan 2.0 / Hunyuan 3.0 / Hy3previewLLM (text)Open-source MoE baseLong-context (256K), agentic coding
HunyuanDiTText-to-imageOpen-sourceChinese-prompt image generation
Hunyuan-VideoText-to-videoOpen-sourceHighest-quality open-source video gen
Hunyuan3D 1.0-2.03D asset generationOpen-sourceSingle image to 3D mesh
HunyuanWorld 1.0 / 1.5 / 2.03D immersive world modelsOpen-sourceExplorable world generation

Six Things the Lineage Tells You

  1. Hunyuan is the most modality-diverse Chinese lineage. Text, image, video, 3D mesh, immersive 3D worlds — all open-source.
  2. The world-model line (HunyuanWorld) is unique. No US lab has shipped an open-source explorable 3D world generator at this scale.
  3. Hy3preview (Apr 2026) introduces fast+slow thinking. Routing between fast inference and deeper reasoning in one model.
  4. 256K context across the LLM line. Consistent long-context across Hunyuan 2.0 and Hy3preview.
  5. WeChat integration is the distribution moat. Hunyuan 3.0 ties to a WeChat AI agent, putting Tencent's AI in front of 1.3B+ MAU.
  6. Hunyuan-Video remains the open-source video benchmark. Most other Chinese labs have not shipped a competitive open video model.

What This Means for AI Visibility

Hunyuan's open-source video and world models matter for brands that produce visual content. As open-source video generation goes mainstream, brand likeness, product appearance, and visual brand identity will increasingly be generated by these models. Auditing how your brand is rendered in Hunyuan-Video and HunyuanWorld is the visual-brand analogue to monitoring how it is described in text.

Methodology

Release dates from Baidu's Tencent HY reference, HunyuanVideo on HuggingFace, HunyuanWorld GitHub, TechNode on Hunyuan 2.0, AiBase on Hunyuan 3.0, and Caixin Global on the Hunyuan 3.0 launch.

How Presenc AI Helps

Presenc AI tracks brand mentions in Tencent Hunyuan-served surfaces including WeChat AI agent answers and Tencent Cloud Hunyuan API outputs, with separate monitoring for the open-source video and world model rendering of brand visual identity.

Frequently Asked Questions

Hy3preview (April 23, 2026): 295B parameters with 21B active and 256K context, combining fast and slow thinking modes. Hunyuan 3.0 is the full launch April 2026, with a major emphasis on coding capability and a tie-in to a WeChat AI agent.
Most of the family is open-source: Hunyuan text-to-image (May 2024), Hunyuan-Large (Nov 2024), Hunyuan-Video (Dec 2024), HunyuanWorld 1.0 / 1.5 / 2.0, Hunyuan3D, and Hy3preview. The latest flagship LLM tier is open-source under permissive terms; some product variants integrated into Tencent Cloud are commercial.
Hunyuan-Video is the leading open-source text-to-video model at 13B parameters, while Sora is proprietary. Hunyuan-Video trails Sora on the highest-quality outputs but is freely usable for commercial work. Most non-Tencent open-source video projects fork from Hunyuan-Video as a base.
It is the first open-source model line that generates explorable, immersive 3D worlds from text or pixels. No other major lab has shipped an open-source equivalent. HunyuanWorld 2.0 (April 2026) is the current state-of-the-art.

Track Your AI Visibility

See how your brand appears across ChatGPT, Claude, Perplexity, and other AI platforms. Start monitoring today.