11-19 更新 实时
首页/ AI 大模型/ 正文

腾讯称混元 Hy4 flash 首日用量超 50 万亿 Token

AI 大模型 来源:TechNode 2026-11-19 15:24 0 阅读 机器翻译
腾讯称混元 Hy4 flash 首日用量超 50 万亿 Token

轻量级 AI 模型的崛起正在挑战“越大越好”这一观念。Tencent 的 Hunyuan 周三表示,Hy4 flash 在 11 月 16 日上线后的首日处理了超过 50 万亿 Token,创下 Hunyuan 系列模型首日使用量的最高纪录。

Hy4 flash 的强劲开局也体现在其在各平台上的表现。在 OpenRouter 的每日 LLM 排行榜上,该模型于 11 月 17 日位列第一。

Tencent 还表示,在其 WorkBuddy 平台上,Hy4 flash 的首日使用量达到其前代产品 Hy4 preview 的 10 倍,并已帮助完成超过 5000 万个真实用户任务。

据 Tencent 介绍,Hy4 flash 采用混合激活架构,总参数量为 2240 亿,其中包含 640 亿 Engram 参数,推理时激活 120 亿参数。

该模型还支持多模态处理,上下文窗口最高可达 100 万 Token。Tencent 表示其生成速度约为每秒 200 Token。但 Hy4 flash 尤其值得关注之处,在于它如何平衡性能与成本。

根据 Tencent 的定价,该模型每百万输入 Token 仅收费 0.5 元(0.07 美元),每百万输出 Token 收费 2 元(0.3 美元),缓存输入 Token 的费用低至每百万 0.02 元(0.003 美元)。

基于 Tencent 内部评估常见办公任务的 WorkIndex 基准测试,Hy4 flash 完成单个任务的成本仅为 Hy4 preview 的十三分之一、Hy4 的二十分之一。其成本也约为同类主流轻量级模型的八分之一。

这些优势使其特别适合高频办公任务,包括创建和编辑演示文稿、文档处理、表格修复、音视频处理以及视觉理解。

此次发布之前,Hy4 已经历过一波需求激增。11 月 4 日该模型在 WorkBuddy 上线时,其智能体能力带来的需求超出预期,许多用户在第一天就用尽了使用额度。腾讯随后将 Hy4 的推理集群扩容十倍,并提高了免费使用额度。

不过这一次,团队似乎已在 Hy4 flash 发布前针对高并发做了更充分的准备,目前尚未出现大范围的使用额度问题。

这一进展也反映出 AI 行业竞争格局的更广泛变化。

随着中国的日均 Token 使用量持续增长,Token 正日益成为衡量 AI 普及程度的重要指标。与此同时,全国各地政府正在推出新举措,从算力券、模型券到基于 Token 的融资,以降低 AI 服务的使用门槛。

竞争不再只是参数规模或模型性能的比拼。在交付出色结果的同时保持低运营成本,正变得同样重要。

Hy4 flash 以 50 万亿 Token 的首次亮相释放出明确信号:市场不仅需要能力更强的 AI 模型,也需要足够便宜、能够支撑日常高频使用的模型。

然而,亮眼的首日使用量与长期商业价值之间仍存在差距。轻量级 AI 模型之间的竞争,可能越来越不只取决于价格和 Token 使用量。

真正的挑战在于持续可靠地交付满足现实业务需求的解决方案,同时保持整体成本低廉。对 Tencent 的 Hunyuan 而言,Hy4 flash 的强势亮相是一个有希望的开端,但其将初期使用势头转化为持续用户价值的能力,将是未来的关键考验。

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Tencent says Hunyuan Hy4 flash surpasses 50 trillion tokens in first-day usage)

The rise of lightweight AI models is challenging the notion that bigger is always better. Tencent’s Hunyuan said on Wednesday that Hy4 flash processed more than 50 trillion tokens on its first day after launching on November 16, marking the highest first-day usage among models in the Hunyuan series.

Hy4 flash’s strong debut was also reflected in its performance across different platforms. On OpenRouter’s daily LLM leaderboard, the model ranked first on November 17.

Tencent also said that on its WorkBuddy platform, Hy4 flash recorded 10 times the first-day usage of its predecessor, Hy4 preview, and had helped complete more than 50 million real-world user tasks.

images created by ChatGPT

According to Tencent, Hy4 flash uses a hybrid activation architecture and has 224 billion parameters in total, including 64 billion Engram parameters, while activating 12 billion parameters during inference.

The model also supports multimodal processing and a context window of up to 1 million tokens. Tencent said its generation speed reaches around 200 tokens per second. But what makes Hy4 flash particularly noteworthy is how it balances performance and cost.

According to Tencent’s pricing, the model charges just 0.5 yuan ($0.07) per million input tokens and 2 yuan ($0.3) per million output tokens, with cached input tokens costing as little as 0.02 yuan ($0.003) per million.

Based on Tencent’s internal WorkIndex benchmark, which evaluates common office tasks, Hy4 flash costs just one-thirteenth as much as Hy4 preview and one-twentieth as much as Hy4 to complete a single task. Its cost is also around one-eighth that of mainstream lightweight models in the same category.

These advantages make it particularly suitable for high-frequency office tasks, including creating and editing presentations, document processing, spreadsheet repair, audio and video processing, and visual understanding.

The launch also comes after an earlier surge in demand for Hy4. When the model debuted on WorkBuddy on November 4, demand driven by its agent capabilities exceeded expectations, with many users hitting their usage limits on the first day. Tencent subsequently expanded its Hy4 inference clusters tenfold and increased free usage quotas.

This time, however, the team appears to have made more extensive preparations for high concurrency ahead of Hy4 flash’s launch, with no widespread usage-limit issues reported so far.

The development also reflects a broader shift in the AI industry’s competitive landscape.

Hy4 flash is now open source.

As daily token usage in China continues to grow, tokens are increasingly becoming an important metric for measuring AI adoption. Meanwhile, local governments across the country are introducing new measures, ranging from computing power and model vouchers to token-based financing, to make AI services more affordable.

Competition is no longer just about parameter counts or model performance. Delivering strong results while keeping operating costs low is becoming equally important.

Hy4 flash’s 50-trillion-token debut sends a clear signal: the market needs not only more capable AI models, but also models that are affordable enough for frequent, everyday use.

However, there is still a gap between strong first-day usage and long-term commercial value. Competition among lightweight AI models may increasingly come down to more than just pricing and token usage.

The real challenge is delivering solutions to real-world business needs consistently and reliably, while keeping overall costs low. For Tencent’s Hunyuan, Hy4 flash’s strong debut is a promising start, but its ability to turn initial usage momentum into sustained user value will be a key test going forward.

正文保留英文原文,标题与摘要由机器翻译。

本文由采集器自动同步

原文发布于 TechNode(Tencent says Hunyuan Hy4 flash surpasses 50 trillion tokens in first-day usage),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗