Tencent Hy4 Flash Lists at $0.07/1M Tokens
Tencent released Hy4 flash on Nov. 16, 2026, a cheaper tier of its Hy4 family with 224 billion total and 12 billion active parameters and a 1M-token context. Tencent Cloud lists it at $0.07 per million input tokens and $0.281 per million output tokens, about 92% and 89% below full Hy4. Tencent's quality and cost claims come from internal tests.
Tencent on Nov. 16 released Hy4 flash, a cheaper and faster tier of its Hy4 large language model family, priced at about $0.07 per million input tokens on Tencent Cloud. The Shenzhen-based company says the model nearly matches its larger predecessor on everyday office work while cutting the cost of a typical task to a twentieth of the full Hy4.

Key takeaways
- Hy4 flash has 224 billion total parameters, including 64 billion Engram parameters, but activates only 12 billion, and accepts up to 1 million tokens of context.
- Tencent Cloud lists it at $0.07 per million input tokens and $0.281 per million output tokens, against $0.834 and $2.501 for full Hy4.
- Tencent's performance and cost claims come from its own internal tests; no independent benchmark results have been published yet.
Hy4 flash price compared with Hy4
Tencent Cloud's TokenHub price list, last updated Nov. 16, puts the new model, called with the identifier hy4-flash, well below the rest of the Hy4 line. Cached input costs $0.003 per million tokens.
| Model | Input | Output | Cache hit |
|---|---|---|---|
| Hy4 flash | 0.07 | 0.281 | 0.003 |
| Hy4 | 0.834 | 2.501 | 0.042 |
| DeepSeek-V4.1-Flash-1017 (peak) | 0.30 | 1.20 | 0.006 |
Based on those list prices, input is about 92% cheaper and output about 89% cheaper than full Hy4. Through 12:00 Beijing time on Nov. 30, TokenHub bills Hy4 flash at half its list price, and Tencent's WorkBuddy assistant offers it free of charge.
Hy4 flash specs: 1M context, 64K output
Tencent Cloud's model list shows a 1-million-token context window, with a maximum input of 960,000 tokens and maximum output of 64,000. The model supports reasoning, structured output, function calling, caching and image understanding.
Tencent attributes the efficiency to three layers of engineering. Hybrid attention, a LatentMoE design and Engram memory shrink the key-value cache to one twenty-fourth of Hy3's and one-sixth of Hy4's. Reinforcement learning cut average end-to-end task output by 50%. Megakernel optimization, 4-bit quantization and a technique Tencent calls DFly push decoding above 150 tokens per second, the company said.
How Tencent measured it
Rather than lead with public leaderboards, Tencent built an internal benchmark called WorkIndex from typical WorkBuddy jobs: creating and editing slide decks, editing documents, repairing spreadsheets, processing audio and video, and visual understanding.
On WorkIndex, Tencent says Hy4 flash slightly outperforms Hy4 Preview, a 770-billion-parameter model with 49 billion active parameters, overall. Its per-task cost is one-thirteenth of Hy4 Preview's, one-twentieth of Hy4's and one-eighth of DeepSeek-V4.1-Flash-1017 at Max-tier peak pricing.
In a separate blind test with 85 internal WorkBuddy users across 259 tasks in 27 categories, Tencent reports user satisfaction close to Anthropic's Claude Haiku 5.5, along with the lowest per-task cost and the shortest completion time among the models it evaluated.
Context: a two-tier Hy4 lineup
Hy4 flash arrives 12 days after Tencent launched full Hy4 on Nov. 4. Tencent says Hy4 ranks second among Chinese models on the Artificial Analysis Intelligence Index and first among open-source models on the Agent Arena leaderboard. TokenHub now marks the older hy4-preview model as due to be retired.
Tencent recommends Hy4 flash for deep research, document processing, slide generation, simple front-end pages and computer use, and full Hy4 for long-horizon software engineering. The split mirrors the flash tiers other labs sell for high-volume agent work, including DeepSeek, whose V4.1 Flash model AI Newsway covered in September.
What's next
Hy4 flash is live in Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis, with OpenRouter support "coming soon." Tencent also says it plans to open-source the model, but has not given a date for either. Independent testing will show whether the internal cost and quality numbers hold outside Tencent's own products.
FAQ
How much does Hy4 flash cost?
On Tencent Cloud TokenHub it lists at $0.07 per million input tokens, $0.281 per million output tokens and $0.003 per million cached input tokens, with a 50% discount through Nov. 30.
Can I download Hy4 flash weights?
Not yet. Tencent says it plans to open-source the model but has not announced a release date.
Should I use Hy4 flash or Hy4?
Tencent recommends Hy4 flash for most everyday tasks and full Hy4 for complex, long-running software engineering.
References
- Tencent Launches Hy4 flash to Power High-Speed, Cost-Efficient AI Deployment for Everyday Workflows — Tencent (Nov. 16, 2026)
- 模型价格 (Model pricing) — Tencent Cloud TokenHub (updated Nov. 16, 2026)
- 模型列表 (Model list) — Tencent Cloud TokenHub (updated Nov. 16, 2026)
- Tencent Launches Hy4 Flash as a Lower-Cost Multimodal Model — Unite.AI (Nov. 16, 2026)
Editor responsible for AI Newsway. Writes about AI, developer tools, cloud and productivity technology.
Comments 0
Comments are currently disabled.