HaiAI123

Curated Global AI Tools Directory

Industry update
Flat

Tencent's Hy4 flash now free on Nous Portal for two weeks

11/16/2026 — 11/17, 04:17·7 sources (all)·7 reports (all)·In the 2026-11-17 briefing·In the 2026-11-16 briefing

Key takeaways

  • On November 16, Tencent released the Hy4 flash language model with 224B total parameters, 12B active, multimodal support, and 1M context, wi…
  • Reddit notes it was not yet open for use. On November 17, Nous Research announced that Hy4 flash is available free on Nous Portal for two we…

Background & analysis

Around 13:09 on November 16, 2026, the Tencent Hunyuan team officially launched the language model Hy4 flash, positioning it as its most cost-effective Hy model to date. The model has 224B total parameters, including 64B Engram parameters, with 12B active parameters per token, and supports multimodal input and a 1M-token context window. API pricing is substantially lower than Hy4's: 0.5 CNY per 1M input tokens and 2 CNY per 1M output tokens, dropping to 0.02 CNY for cached input. Official material attributes the cost efficiency to coordinated gains across pretraining, post-training, and inference. Pretraining adopts hybrid attention, LatentMoE, and Engram architectures, cutting KV cache to 1/24 of Hy3's and 1/6 of Hy4's. Post-training self-distills Hy4 and uses reinforcement learning to shorten end-to-end task output length by 50%. Inference combines megakernel, 4bit quantization, and dflash, pushing decode speed above 150 tps.

Through the afternoon, independent outlets shared early results. Zhidongxi used the model to write code that produced a pixel-style MV of Jensen Huang, reported the input and output API prices at 1/12 and 1/9 of Hunyuan 4's full release respectively, and noted average per-task cost of about 0.2 CNY on Tencent's everyday-task evaluation. ifanr said roughly 0.2 CNY was enough to generate a 3D interactive web page from scratch and judged it more cost-effective than DeepSeek V4.1 Flash, without offering systematic benchmark comparisons. Reports diverged on official per-task cost claims: IT Home cited roughly 5% of Hy4's cost on the WorkIndex set, while InfoQ cited 1/20 of Hy4 and 1/13 of Hy4 preview. A Reddit LocalLLaMA post quoting official material said the model was not yet open for general use, while InfoQ reported it was already available on Tencent Cloud TokenHub, Yuanbao, and WeCom.

On November 17, 2026, Nous Research announced on X that Hy4 flash from @TencentHunyuan would be free in Nous Portal for two weeks, highlighting its cost-effectiveness on everyday work tasks such as documents, slides, and computer use. That free two-week window is where the event currently stands.

AI-generated from 7 reports · updated 7 hours ago

Latest turnNous Research says Hy4 flash, a new model from Tencent Hunyuan, is free on Nous Portal for two weeks. It is built to be extremely cost-effective at everyday work tasks such as documents, slides and computer use.

Related AI tools

updream
24-hour heatHeat index 431 · peak 536 · 7h ago
24 hours agonow

Reports on this story headlines open the original

Today
  1. Nous Research says Hy4 flash, a new model from Tencent Hunyuan, is free on Nous Portal for two weeks. It is built to be extremely cost-effective at everyday work tasks such as documents, slides and computer use.

    X:Nous Research (@NousResearch)AI score 61
Yesterday
  1. Tencent Hunyuan released Hy4 flash, a 224B-parameter model with 12B active parameters that supports multimodal input and a 1M context window, priced at 0.5 yuan per million input tokens and 2 yuan per million output tokens, or 0.02 yuan on cache hits. Tencent says internal WorkIndex testing puts its per-task cost at roughly 1/20 of Hy4's, and it is already live across TokenHub, Yuanbao and WeCom.

    InfoQ 中文AI score 57
  2. iFanr put Tencent's Hy4 flash through a hands-on test, saying about 0.2 yuan was enough to have the model build a 3D interactive webpage from scratch. The outlet calls it better value than DeepSeek V4.1 Flash, based on its own generated output rather than a systematic benchmark.

    爱范儿AI score 53
  3. Tencent introduced Hy4 flash with 224B total parameters, including 64B Engram parameters, and just 12B active per token, according to an official blurb relayed on Reddit's LocalLLaMA. It is natively multimodal with a 1M-token context window and significantly lower API pricing, though the post suggests it has not yet been released.

    Reddit · LocalLLaMAAI score 54
  4. Tencent has launched Hunyuan 4 flash (Hy4 flash), a model with 224B total parameters, 12B activated parameters, multimodal support, a 1M-token context window, and over 150 tokens/s decoding. Its API costs CNY 0.5 per 1M input tokens and CNY 2 per 1M output tokens, 1/12 and 1/9 of the standard Hunyuan 4 rates; Zhidongxi tested it by coding a pixel-art Jensen Huang music video.

    智东西AI score 68
  5. On November 16, Tencent Hunyuan released Hy4 flash, a language model with 224B total parameters (including 64B Engram parameters) and 12B active parameters. It supports multimodal input and a 1M context window, with pricing at CNY 0.5 per million input tokens and CNY 2 per million output tokens.

    36氪AI score 67
  6. Tencent's Hunyuan team released Hy4 flash, a 224B-parameter model with 12B active per token, multimodal input and a 1M-token context window. Tencent says per-task cost on its WorkIndex benchmark is about 5% of Hy4's, with API pricing cut sharply as a result.

    IT之家AI score 62

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 →