---
title: "Daily average of 30 trillion Tokens! The comeback of \"Doubao\""
description: "ByteDance's \"Doubao\" large model has achieved an average daily token call volume of 30 trillion, realizing a 253-fold increase, with 157 million monthly active users surpassing DeepSeek, ranking first"
type: "news"
locale: "en"
url: "https://longbridge.com/en/news/261561584.md"
published_at: "2025-10-17T06:54:18.000Z"
---

# Daily average of 30 trillion Tokens! The comeback of "Doubao"

> ByteDance's "Doubao" large model has achieved an average daily token call volume of 30 trillion, realizing a 253-fold increase, with 157 million monthly active users surpassing DeepSeek, ranking first among AI applications in China. The key to its success lies in its deep integration with Douyin, rich multimodal capabilities, and low-threshold user experience, proving that in the AI market, product capabilities and ecosystem integration are more important than pure technological advantages

ByteDance's large model application "Doubao" is achieving a dual breakthrough in users and technology at an astonishing speed. Launched in 2023, this AI assistant has surpassed DeepSeek in the fiercely competitive Chinese AI market, becoming the AI application with the most monthly active users, thanks to its differentiated product strategy and ByteDance's traffic distribution capabilities.

On October 16, Tan Dai, president of Volcano Engine, disclosed at the FORCE LINK AI Innovation Tour held in Wuhan that **the daily Token call volume of the Doubao large model has increased from 120 billion in May last year to 30 trillion in September this year, achieving a 253-fold growth**. According to an IDC report, in the first half of this year, Volcano Engine ranked first in the call volume of large model services in China's public cloud, with a market share of 49.2%. QuestMobile data shows that as of August, Doubao's monthly active users reached 157 million, surpassing DeepSeek's 143 million.

To lower the cost threshold for enterprises to apply AI, **Volcano Engine upgraded the Doubao large model to version 1.6, launching the country's first model that natively supports four levels of thinking length adjustment, and released an intelligent model routing solution**. In low-intensity mode, the new version can reduce total output Tokens by 77.5% and deep thinking time by 84.6%. The intelligent routing solution can reduce overall costs by more than 70% in cost-priority mode.

Analysts point out that this turnaround highlights the differences in technological leadership and product capabilities. **Doubao, relying on ByteDance's "application factory" experience and deep integration with Douyin, is proving that in the consumer-grade AI market, user experience and distribution capabilities are as important as model performance itself**.

## Tiered Thinking Solves Cost Dilemma

This upgrade of Doubao directly addresses the core pain point of enterprise AI applications—excessively high inference costs.

The Doubao large model 1.6, released by Volcano Engine in Wuhan, natively supports four thinking lengths: Minimal, Low, Medium, and High, allowing enterprises to flexibly adjust according to different scenario needs.

Tan Dai stated that this is one of the three major trends in the AI field this year— the emergence of deep thinking models—giving rise to this solution.

> The Doubao-Seed-1.6, launched in June this year with multimodal understanding capabilities, improved model performance by 31% after entering thinking mode, but in real-time scenarios, the proportion of thinking Tokens exceeded 77%, leading 82% of enterprises to choose not to enable deep thinking mode.
> 
> The upgraded Doubao-Seed-1.6-1015 new version can reduce total output Tokens by 77.5% and deep thinking time by 84.6% while maintaining the same effect through tiered adjustment in low-intensity mode.

The simultaneously released Doubao-Seed-1.6-lite model is smaller in size and outperforms the previous Pro version, with total costs reduced by 53% compared to Doubao-Seed-1.5-Pro within the 0-32K context range Volcano Engine has also launched an intelligent model routing solution that supports balanced mode, effect priority mode, and cost priority mode, allowing for automatic selection among the entire series of Doubao models as well as mainstream open-source models like DeepSeek, Qwen, and Kimi.

## User Experience-Driven Differentiated Competition

Doubao's market leadership stems more from its product strategy than purely from technological advantages.

According to technology analyst Poe Zhao, **DeepSeek is more like "the first generation of AI applications, similar to early ChatGPT,"** while Doubao represents China's second-generation AI platform, **"integrating richer features, clear visual prompts, and scenario-based guidance, significantly lowering the cognitive threshold for using large models."**

Doubao integrates various functions such as chat, image generation, video generation, and custom intelligent agents, most importantly, it is deeply integrated with Douyin.

Douyin users can tag Doubao in video comments to obtain content summaries, and Doubao users can also watch Douyin videos directly within the app. Counterpoint Research's chief AI analyst Wei Sun pointed out:

> "This combination of convenience and dopamine, especially for the younger generation, is just enough to trigger sharing. This lightweight, low-threshold user experience is key to driving social sharing and sticky interaction in China."

QuestMobile data shows that as of August, nearly 40% of users leaving DeepSeek turned to Doubao.

## Building an Enterprise-Level AI Ecosystem

To promote enterprise AI transformation, Volcano Engine is improving its full-chain solutions from development to deployment.

In terms of Agent development, Volcano Engine has built a "1+N+X" system through the HiAgent platform, which includes one unified interaction entry, N general intelligent assistants for enterprises, and X custom intelligent agent applications for enterprises.

Volcano Engine plans to launch a corporate version of Kousi Space, providing customized Agents, MCP development, data connectors, enterprise knowledge bases, and enterprise-level system management functions to meet personalized enterprise needs.

Reports indicate that ByteDance has collaborated with device OEM manufacturers, including smart glasses manufacturers and automotive manufacturers (including Tesla), to embed Doubao as an in-car assistant or AI companion, expanding Doubao's application range from smartphones and computers to more terminals.

## From Technological Competition to Application Implementation

The rise of Doubao reveals a shift in the competitive landscape of China's AI market.

DeepSeek caused a sensation globally in January this year due to its efficient model, but its product remains at the basic text chat stage, lagging in multimodal functionality and user guidance. Dermot McGrath pointed out:

> Some users left DeepSeek in search of image generation capabilities or autonomous intelligent agents, while others left because DeepSeek's servers often could not handle traffic In contrast, ByteDance is known as the "application factory" in China, as it regularly launches dozens of new applications for various use cases.

> ByteDance has strong capabilities in rapid iteration, quick testing, and traffic diversion between applications. This long-term accumulation in consumer-grade platform development is key to Doubao surpassing the technology-leading but product-experienced DeepSeek.

Silicon Valley's top venture capital firm a16z ranked **Doubao as the fourth most popular generative AI application globally in August, following ChatGPT and Google's Gemini**.

Analysis indicates that this application, which has almost no recognition in the West, is proving with a growth rate of 253 times that in the AI race, technological leadership is just the starting point; product capability and ecosystem integration are the keys to determining market position

### Related Stocks

- [BYTED.NA - ByteDance](https://longbridge.com/en/quote/BYTED.NA.md)

## Related News & Research

| Title | Description | URL |
|-------|-------------|-----|
| AI 影片生成模型 Seedance 惹議 好萊塢控嚴重侵權 | 字節跳動推出的 AI 影片生成模型 Seedance 2.0 引發爭議，美國電影協會指控其未經授權大規模使用受著作權保護的作品。該模型生成的影片在網絡上廣泛傳播，包括湯姆克魯斯與布萊德彼特的打鬥場景。電影協會主席李夫金表示，Seedance | [Link](https://longbridge.com/en/news/275949589.md) |
| 對 Seedance2.0 們的倫理敲打正當時 | 字節的 Seedance 2.0 視頻生成模型在發佈後迅速引起關注，展現出超越玩具階段的能力。儘管發佈後不久，字節對其生成能力進行了緊急收緊，限制了真人生成和名人素材的使用。這一調整反映了 AI 視頻模型在能力與現有規則之間的矛盾。影視颶風 | [Link](https://longbridge.com/en/news/275337447.md) |
| 字節跳動自研 AI 晶片，三星談代工今年目標 10 萬顆 | 字節跳動正在研發自有 AI 晶片，並與三星電子洽談代工，以確保處理器供應穩定。由於美國出口管制，字節跳動啟動了代號為「SeedChip」的晶片計劃，旨在保障 AI 算力供應。預計在今年 3 月底前取得首枚樣品，並計劃今年生產至少 10 萬枚 | [Link](https://longbridge.com/en/news/275713889.md) |
| 歐盟公佈初步調查 指 TikTok 違反數位服務法 | 歐盟今日表示，TikTok 違反數位服務法，未能有效解決其設計對用户，尤其是年輕人和兒童的負面影響。歐盟執行委員會要求 TikTok 改善其「令人成癮的設計」，如無限捲動和個性化推薦系統，若不改進將面臨高額罰款。調查顯示，TikTok 對成 | [Link](https://longbridge.com/en/news/275137226.md) |
| TikTok 在美成立合資企業後 用户卸載大增近 150% 創作者轟審查機制失控 | TikTok 在美國成立合資企業後，用户卸載應用程式的速度顯著加快，過去 5 天內每日卸載量較前三個月增加近 150%。用户對新隱私政策表示擔憂，部分創作者在社交媒體上表達了對合資企業的懷疑，甚至刪除了自己的 TikTok 帳號。新政策涉及 | [Link](https://longbridge.com/en/news/273784234.md) |

---

> **Disclaimer**: This article is for reference only and does not constitute any investment advice.