HaiAI123

Curated Global AI Tools Directory

Industry update
↓ 24%

Tencent Hunyuan open-sources Hy4, releasing its full LLM lineup

11/19/2026 — 11/20, 16:19·5 sources (all)·5 reports (all)·In the 2026-11-20 briefing

Key takeaways

  • Tencent Hunyuan has open-sourced Hy4, with weights released on GitHub, Hugging Face, ModelScope, and GitCode under the Apache 2.0 license. T…

Background & analysis

On November 4, 2026, Tencent Hunyuan released the official version of Hy4 and signaled that the model weights would be opened up at a later date. On November 19, it launched Hy4 flash, a lightweight multimodal model with 224B total parameters, including 64B of Engram parameters, and 12B active parameters per token. It supports a 1M context window and text and image inputs. Pricing was set at CNY 0.5 per 1M tokens for input and CNY 2 for output, dropping to as low as CNY 0.02 per 1M tokens when the cache hits. The cost per task was CNY 0.26, or 1/13 that of Hy4 preview and 1/20 that of Hy4. According to InfoQ Chinese, Hy4 flash topped the OpenRouter LLM daily ranking on its first day, with token calls surpassing 50 trillion, and it is now available on Tencent Cloud TokenHub. Tencent Hunyuan described it as the most cost-effective Hy model so far.

On November 20, Together AI announced that Tencent Hunyuan's hy4 had gone live on its platform, describing it as a multimodal model built for long-horizon agentic work, including coding and computer use, with native vision and a 1M context window. The same day, Tencent Hunyuan formally open-sourced Hy4, with model weights posted simultaneously on GitHub, Hugging Face, ModelScope, and GitCode, along with a Hy4 System Card laying out the safety evaluation methods, test results, and known risks. The release package also covers an FP8 quantized version and a pretrained base model.

IT之家 and cnBeta give matching specifications: Hy4 uses a MoE architecture, with 770B total parameters and roughly 49B active parameters per token, and a context window of up to 1M tokens, supporting text, image, and video understanding. Hy4 and Hy4 preview are released under the Apache 2.0 license, allowing commercial use and modification. For Hy4 flash, InfoQ Chinese lists total parameters of 224B, while IT之家 describes roughly 220 billion, a discrepancy that remains unresolved. With Hy4 now available, the full Hy lineup, including Hy3 preview, Hy3, Hy4 preview, Hy4, and Hy4 flash, has been open-sourced and is downloadable for self-hosted deployment.

AI-generated from 5 reports · updated 2 days ago

Latest turnTencent Hunyuan released Hy4 as an open-source model on November 20, alongside a System Card detailing its safety evaluation methods and known risks, according to cnBeta. Hy4 is a mixture-of-experts multimodal model with 770B total parameters, roughly 49B activated per token, a context window of up to 1M tokens, and support for text, image, and video understanding; FP8-quantized and pretrained base versions were also released. With Hy4 out, the full Hunyuan Hy LLM series is now open source.

Related AI tools

updream
24-hour heatHeat index 81 · peak 244 · 23h ago
24 hours agonow

Reports on this story headlines open the original

Nov 20
  1. Tencent Hunyuan released Hy4 as an open-source model on November 20, alongside a System Card detailing its safety evaluation methods and known risks, according to cnBeta. Hy4 is a mixture-of-experts multimodal model with 770B total parameters, roughly 49B activated per token, a context window of up to 1M tokens, and support for text, image, and video understanding; FP8-quantized and pretrained base versions were also released. With Hy4 out, the full Hunyuan Hy LLM series is now open source.

    cnBetaAI score 72
  2. Tencent Hunyuan has open-sourced its Hy4 model, available for download and deployment on GitHub, Hugging Face, ModelScope and GitCode. The company says its full LLM lineup is now open source, covering Hy3 preview, Hy3, Hy4 preview, Hy4 and Hy4 flash.

    36氪AI score 67
  3. Tencent's Hunyuan team has open-sourced the Hy4 model, with weights available on GitHub, Hugging Face, ModelScope and GitCode under Apache 2.0. Hy4 uses a MoE architecture with 770B total parameters and 49B active parameters, and supports a context window of over 1 million tokens; the release completes the full open-sourcing of the Hunyuan LLM series.

    IT之家AI score 74
  4. Together AI announced that Tencent Hunyuan's hy4 is now live on its platform and available to try. The multimodal model targets long-horizon agentic work, including coding and computer use, with native vision and a 1M-token context window.

    X:Together AI (@togethercompute)First-partyAI score 46
Nov 19
  1. Tencent Hunyuan released Hy4 flash, a lightweight multimodal model with 224B total and 12B active parameters and a 1M-token context, priced at 0.5 yuan per million input tokens and 2 yuan for output, dropping to 0.02 yuan on cache hits. It drew 50 trillion tokens of calls on its first day, topped OpenRouter's daily leaderboard and is now available on Tencent Cloud TokenHub, InfoQ China reported.

    InfoQ 中文AI score 72

Other stories people are talking about

How is heat calculated?About the method

Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.

This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.

Surge
Discussion rising fast
New
First report within 6 hours
Rising
Still gathering discussion

Back to AI industry updates →