Tencent 推出 Hy4 flash,主打更省錢地處理日常工作
Tencent Hy4 Flash
日常工作常要在成果品質、完成時間和費用之間取捨。
- Tencent 推出 Hy4 flash,這款多模態模型每次處理 token 時啟用 12B 引數,並提供 1M-token 上下文視窗。Tencent 表示,在 WorkBuddy 評估中,它的日常任務品質接近 Hy4,平均完成時間少了 70%,費用約為 Hy4 的 1/15。
- 但處理需要長時間投入的程式開發工作時,Hy4 的能力上限較高。Hy4 flash 已在多項 Tencent 服務提供,Tencent 也計畫推出開放權重版本。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
November 16, 2026
Introducing Hy4 flash: Ultimate cost-efficiency for everyday productivity
We’re introducing Hy4 flash, our most cost-effective Hy model to date—featuring 224B total parameters (including 64B Engram parameters) with just 12B active parameters per token. Like Hy4, it is natively multimodal with a 1M-token context window, while offering significantly lower API pricing.
| API pricing (per million tokens) | Cached input | Input | Output |
|---|---|---|---|
| Hy4 flash | $0.003 | $0.070 | $0.281 |
| Hy4 | $0.042 | $0.834 | $2.501 |
Across training, evaluation, and inference, we aggressively optimized for cost efficiency while preserving intelligence:
- Pre-training: Hy4 flash integrates hybrid attention, LatentMoE, and Engram into its architecture—reducing KV cache size to just 1/24 of Hy3 and 1/6 of Hy4, with extensive scaling ladder experiments ensuring performance meets expectations.
- Post-training: Self-distillation from Hy4 establishes strong general intelligence, while reinforcement learning cuts average end-to-end output tokens by 50%.
- Evaluation: We built WorkIndex, an internal benchmark grounded in typical WorkBuddy tasks, and paired it with human evaluation—focusing on deliverable quality, single-turn output length, and task turnaround time rather than public benchmark scores detached from everyday work.
- Inference: Combining megakernels, 4-bit quantization, and DFly pushes decoding speed past 150 tokens/s, enabling low-cost serving at scale.
Together, these gains bring Hy4 flash slightly above Hy4 preview in terms of overall task performance. On WorkIndex, its average cost per task is just 1/13 of Hy4 preview, 1/20 of Hy4, and 1/8 of DeepSeek-V4.1-Flash-1017 (Max tier, peak-hour pricing).
Figure 1: Performance and cost across internal and external tasks. WorkIndex comprises 8 evaluation sets covering slide creation, slide editing, human-AI slide co-writing, Markdown human-AI co-writing, document editing, spreadsheet repair, audio/video processing, and visual understanding. Open Benchmarks includes 7 public agent benchmarks: DeepSWE, ProgramBench, Terminal-Bench 4.0, APEX-Agents, AutomationBench, HLE, and OSWorld 2.1.

Figure 2: Average cost per task on WorkIndex.
In a blind evaluation with 85 internal WorkBuddy users across 259 tasks spanning 27 categories, Hy4 flash achieved user satisfaction close to Claude Haiku 5.5, while recording the shortest end-to-end completion time and significantly lower cost per task than any other evaluated model.

Hy4 flash is now available across Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima, Marvis, and more, and will launch on OpenRouter soon. Through November 30, users can access it for free in WorkBuddy and get 50% off API pricing via Tencent Cloud TokenHub.
We recommend Hy4 flash for most everyday productivity tasks—such as deep search, document processing, slide creation, simple frontend demos, and computer use—while continuing to recommend Hy4 for complex workloads like long-horizon software engineering for its higher intelligence ceiling. We also plan to release Hy4 flash as an open-weight model and ship frequent updates to continuously improve its utility and efficiency across real-world workflows.

In everyday office work, users care just as much about turnaround time and cost as they do about execution quality. In WorkBuddy’s evaluations, Hy4 flash cuts average task time by 70% at roughly 1/15 the cost of Hy4—matching Hy4’s quality on everyday tasks with only a slight dip on long-horizon coding workloads. Combined with its native multimodal capabilities, Hy4 flash is our go-to choice for balancing quality, speed, and cost.

In Yuanbao workflows, Hy4 flash boosts agent performance by 13.6% over Hy3 and achieves a 7.56% gain in human side-by-side (GSB) agent evaluations—standing out for more thorough task completion, clearer information presentation, and more credible, professional outputs. On general chat tasks, it also outperforms GLM-5.3-Flash by 2.52%, with the most notable gains in multi-turn dialogue.
Yuanbao Logan, Product Lead
[](https://hy.tencent.ai/research/hy4-flash#appendix-benchmark-results)Appendix: Benchmark results
