Artificial intelligence news.
Straight from the original announcements.

Tencent launches Hy4 flash for lower-cost AI work

November 17, 2026, 10:11By the AI News Bites Editorial TeamSource: PR Newswire
Tencent launches Hy4 flash for lower-cost AI work
Image: PR Newswire

Tencent has released Hy4 flash, a multimodal AI model designed for everyday productivity workflows with up to 1M tokens of context and 12 billion active parameters. The company says it delivers lower deployment costs and faster task completion than its larger Hy4 models.

Key points
  • Hy4 flash has 224 billion total parameters and 12 billion active parameters.
  • It supports multimodal inputs and context lengths of up to 1M tokens.
  • Tencent lists prices from approximately US$0.003 per million cached input tokens.
  • The model is available across Tencent's listed services, with OpenRouter integration planned.
  • Tencent plans to open-source Hy4 flash with frequent iterations.

A smaller active model with a long context window

Hy4 flash is the latest addition to Tencent's Hy model family. It has 224 billion total parameters, including 64 billion Engram parameters, while using only 12 billion active parameters. The model supports multimodal inputs and a context length of up to 1M tokens.

Tencent describes Hy4 flash as its most cost-effective Hy model to date. The company lists pricing of approximately US$0.070 per million input tokens, US$0.281 per million output tokens, and US$0.003 per million cached input tokens. Tencent says these prices represent a substantial reduction from Hy4.

The model uses hybrid attention, LatentMoE, and Engram features during pre-training. Tencent also says its KV cache is one 24th the size of Hy3's and one sixth the size of Hy4's. The company conducted scaling ladder experiments as part of its development process.

Engineering choices target speed and efficiency

For post-training, Hy4 flash draws on Hy4's architecture. Tencent says reinforcement learning reduces average end-to-end task output tokens by 50 percent. Rather than relying only on public benchmark scores, the company created WorkIndex, an internal benchmark based on typical WorkBuddy tasks.

WorkIndex measures deliverable quality, single-turn output length, and task turnaround time. Its tasks include PowerPoint deck creation and editing, document editing, spreadsheet repair, audio and video processing, and visual understanding.

At inference, Tencent combines megakernel optimization, 4-bit quantization, and DFly techniques. According to the company, Hy4 flash reaches decoding speeds exceeding 150 tokens per second. Tencent says the model slightly outperforms the larger Hy4 Preview in overall effectiveness. Hy4 Preview has 770 billion total parameters and 49 billion active parameters.

Tencent positions Hy4 flash for everyday work

Based on WorkIndex, Tencent says Hy4 flash has a per-task cost of 1/13 that of Hy4 Preview, 1/20 that of Hy4, and 1/8 that of DeepSeek-V4.1-Flash-1017 at the Max tier and peak-hour pricing. In a blind evaluation involving 85 internal Tencent WorkBuddy users, 259 tasks, and 27 categories, the company says user satisfaction was close to Claude Haiku 5.5. Tencent also says Hy4 flash had the lowest per-task cost among the evaluated models and the shortest task completion time.

Hy4 flash is available through Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima, and Marvis. OpenRouter integration is planned. Users can access the model for free on WorkBuddy through November 30 and receive 50 percent off standard pricing through Tencent Cloud TokenHub.

Tencent recommends Hy4 flash for deep research, document processing, PowerPoint generation, simple front-end displays, and computer use. It directs users to Hy4 for complex tasks such as long-horizon software engineering. Tencent also plans to open-source Hy4 flash and release frequent iterations, extending the model's availability while focusing on practical utility and efficiency for real-world work.

TencentHy4 flashHy4WorkBuddyTencent Cloud TokenHubDeepSeek-V4.1-Flash-1017language modelsmultimodal AIenterprise AIbenchmarks

Related news

Tencent Hy4 goes live on Together AI with 1M-token context and image input
ModelsNovember 19, 2026

Tencent Hy4 goes live on Together AI with 1M-token context and image input

Together AI has made Tencent's flagship Hy4 model available through serverless and dedicated endpoints. Hy4 is the first model in Tencent's Hy family to accept images…

Tencent Hy4 goes GA on Novita AI at the same price as Preview
ModelsNovember 4, 2026, 11:00

Tencent Hy4 goes GA on Novita AI at the same price as Preview

Novita AI has made Tencent Hy4 its official GA model under the ID tencent/hy4, keeping the price identical to Hy4 Preview while adding image input to the catalog entry.…

Vercel adds Tencent Hy4 to AI Gateway with 50% discount
ModelsNovember 3, 2026

Vercel adds Tencent Hy4 to AI Gateway with 50% discount

Vercel has made Tencent’s Hy4 model available through AI Gateway, offering 50% off through December 31, 2026. The model supports vision, reasoning, tool calling…