# Doubao

> Source: https://aiwiki.ai/wiki/doubao
> Updated: 2026-11-01
> Fact-checked: 2026-11-02
> Categories: Chinese AI, Conversational AI, Large Language Models
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Doubao." aiwiki.ai, 1 Nov 2026. https://aiwiki.ai/wiki/doubao
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Doubao** (豆包, literally "bean bun") is China's most-used [artificial intelligence](https://aiwiki.ai/wiki/artificial_intelligence) [chatbot](https://aiwiki.ai/wiki/chatbot), developed by [ByteDance](https://aiwiki.ai/wiki/bytedance_ai), the parent company of [TikTok](https://aiwiki.ai/wiki/tiktok). Launched in August 2023, Doubao surpassed 100 million daily active users (DAU) in December 2025, and its DAU spiked to over 100 million in a single day again during the 2026 Lunar New Year period after ByteDance's Volcano Engine served as the exclusive AI cloud partner of the CCTV Spring Festival Gala.[12][17][18] At the peak of that broadcast, Doubao processed 63.3 billion tokens in a single minute.[18] The app is powered by [ByteDance](https://aiwiki.ai/wiki/bytedance)'s proprietary Seed model family and is available on iOS, Android, web, and desktop platforms. Its international counterpart, originally released as Cici and later rebranded to Dola, serves users in select overseas markets.[23][40]

<figure class="w-56">
<img src="/images/wd-2026-10-28-bytedance-logo-f60fd0c64e09-tile.png" alt="ByteDance logo" width="480" height="138" />
<figcaption>ByteDance logo. Source: <a href="https://commons.wikimedia.org/wiki/File:ByteDance_logo_English.svg">Wikimedia Commons</a>, <a href="https://commons.wikimedia.org/wiki/File:ByteDance_logo_English.svg#Licensing">Public domain</a>.</figcaption>
</figure>

Doubao offers a wide range of capabilities including text-based conversation, image generation (via the [Seedream](https://aiwiki.ai/wiki/seedream) model), video generation (via the [Seedance](https://aiwiki.ai/wiki/seedance) model), document analysis, voice interaction, and code generation.[42] Through ByteDance's [Volcengine](https://aiwiki.ai/wiki/volcano_engine) cloud platform, the underlying Doubao models are also available as enterprise APIs, where aggressive pricing (as low as 99.3% below industry averages at launch) triggered a price war across China's AI sector in 2024.[2]

## What is Doubao used for?

Doubao functions as a free, general-purpose AI assistant for Chinese consumers and as a commercial Model-as-a-Service API for enterprises. Consumers use it for everyday question-and-answer, writing and brainstorming, document and chart analysis, real-time voice conversation, and creative image and video generation, all in a single app. Enterprises consume the same underlying Seed models through Volcengine to build their own products, with daily token consumption on the platform exceeding 50 trillion tokens by December 2025, a 417-fold increase since the model family launched in May 2024.[10][50] Doubao was free for consumers in China until June 24, 2026, when ByteDance introduced paid professional plans alongside the free tier.[47][48]

## History

### Origins and Early Development (2023)

ByteDance began internal testing of an AI chatbot in June 2023, initially under the codename "Yunque." The project was developed by ByteDance's Seed research team, which was established in early 2023 as the company's response to the rapid advancements in [large language models](https://aiwiki.ai/wiki/large_language_model) following the release of [ChatGPT](https://aiwiki.ai/wiki/chatgpt) in November 2022. ByteDance's earlier AI Lab, founded in 2016, was later fully merged into the Seed team to consolidate AI research efforts.

On August 17, 2023, the chatbot launched publicly under the name "Doubao" and was open to users in China without an invite code.[1] The underlying [language model](https://aiwiki.ai/wiki/language_model) was initially called Skylark (also referred to as the "Lark Model") before being rebranded under the Doubao name. At the same time, ByteDance released an international version of the chatbot called Cici, targeting overseas markets through a Singapore-based subsidiary, SPRING (SG) PTE. LTD.[23]

Doubao was among the first [AI models](https://aiwiki.ai/wiki/model) in China to receive algorithm registration from Chinese regulators, allowing it to operate as a consumer-facing product under the country's [generative AI](https://aiwiki.ai/wiki/generative_ai) regulations.[2][3] The Cyberspace Administration of China (CAC) approved the launch of Doubao alongside [Baidu](https://aiwiki.ai/wiki/baidu_ai)'s Ernie Bot at the end of August 2023, shortly after the country's [interim Measures for the Management of Generative AI Services](https://aiwiki.ai/wiki/china_interim_measures) took effect on August 15, 2023.

### Rapid Growth and Price War (2024)

In May 2024, ByteDance formally launched the Doubao model family through its Volcengine cloud platform.[3] The company set prices at 0.0008 yuan (roughly 0.011 US cents) per 1,000 tokens for its enterprise-tier model.[3] According to Tan Dai, president of Volcengine, this was 99.3% less than the industry average of approximately 12 Chinese cents per 1,000 tokens for models of equivalent specification.[2] The aggressive pricing strategy immediately triggered a broader price war across China's enterprise large language model sector, forcing competitors such as [Alibaba](https://aiwiki.ai/wiki/alibaba_cloud), [Baidu](https://aiwiki.ai/wiki/baidu), and [Tencent](https://aiwiki.ai/wiki/tencent) to slash their own API fees within days.[2]

By May 2024, the Doubao app had passed 100 million downloads and had about 26 million monthly users.[3] By November 2024, it had nearly 60 million monthly active users, which Volcano Engine said made it second only to ChatGPT among AI apps.[55]

At the December 2024 Volcengine FORCE conference (Winter edition), ByteDance announced a comprehensive upgrade to the Doubao model family. Key releases included the Doubao General Model Pro (with 32% improved task-handling ability compared to May), a visual understanding model priced at just 0.003 yuan per 1,000 tokens (85% below the industry average), and a 3D generation model capable of producing high-fidelity 3D assets within one minute. ByteDance also announced that the Doubao video generation model would officially open for external services in January 2025.

### Scaling to 100 Million DAU (2025)

The year 2025 saw Doubao's usage metrics accelerate dramatically. In January 2025, ByteDance released the Doubao-1.5-Pro model, a sparse [Mixture of Experts](https://aiwiki.ai/wiki/mixture_of_experts) (MoE) architecture with 20 billion activated parameters that matched the performance of a 140-billion-parameter dense model.[4] The model came in two context-length configurations (32K and 256K tokens) and featured a "deep thinking" reasoning mode.[4] [Benchmarks](https://aiwiki.ai/wiki/benchmark) put it on par with [GPT-4o](https://aiwiki.ai/wiki/gpt-4) and [Claude 3.5 Sonnet](https://aiwiki.ai/wiki/claude), and ahead of DeepSeek-V3 and Llama 3.1 405B on tests including AIME.[4]

In April 2025, ByteDance disclosed the technical details for **Seed-Thinking-v1.5**, a Mixture-of-Experts reasoning model with 20 billion activated parameters and 200 billion total parameters.[5] The team described it succinctly in the technical report: "Seed1.5-Thinking is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters."[5] The model was trained with a custom [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning) stack featuring two ByteDance-developed algorithms, VAPO (a value-based actor-critic method) and DAPO (a policy-gradient method), designed to stabilize long [chain-of-thought](https://aiwiki.ai/wiki/chain_of_thought) training and reduce reward sparsity.[5] Seed-Thinking-v1.5 reported 86.7 on [AIME 2024](https://aiwiki.ai/wiki/aime_2024), 55.0 on Codeforces, and 77.3 on GPQA, and "surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability."[5] [Volcano Engine](https://aiwiki.ai/wiki/volcano_engine) launched the related Doubao 1.5 deep-thinking model at a FORCE LINK roadshow on April 17, 2025.[53]

In May 2025, ByteDance published the Seed1.5-VL technical report and released the model on Volcengine.[7] This [vision-language model](https://aiwiki.ai/wiki/multimodal_model), composed of a 532-million-parameter [vision encoder](https://aiwiki.ai/wiki/vision_encoder) and an MoE LLM with 20 billion active parameters, achieved state-of-the-art performance on 38 out of 60 public benchmarks.[7] In agent-centric tasks such as GUI control and gameplay, Seed1.5-VL outperformed leading multimodal systems including [OpenAI](https://aiwiki.ai/wiki/openai) CUA and Claude 3.7.[7]

By August 2025, according to QuestMobile's AI Application Industry Monthly Report, Doubao reached 157 million monthly active users with a 6.6% month-over-month growth rate, officially surpassing [DeepSeek](https://aiwiki.ai/wiki/deepseek) to claim the top position among native AI applications in China.[8] Over the same period DeepSeek's chatbot declined 4% to roughly 143 million MAU, and QuestMobile estimated that nearly 40% of users who left DeepSeek in May migrated to Doubao.[8][49]

In October 2025, Volcengine launched Doubao 1.6-Vision, the first model in the family to use tool calling (such as cropping, scaling and annotating images) for visual reasoning, at about 50% lower cost than its predecessor, Doubao-1.5-Thinking-Vision-Pro.[9]

In December 2025, at the Volcengine FORCE conference, ByteDance launched the Doubao Large Model 1.8 alongside the Seedance 1.5 Pro video generation model.[10] The 1.8 model was optimized specifically for multimodal agent scenarios, with enhanced tool invocation, complex instruction compliance, and OS agent capabilities.[11] ByteDance reported a score of 67.6 on the BrowseComp-en [agent evaluation](https://aiwiki.ai/wiki/agent_evaluation) benchmark, ahead of Gemini 3 Pro.[11] At this point, Doubao's daily average token consumption exceeded 50 trillion, representing a 417-fold increase since May 2024.[10][50] More than 100 enterprise customers had each accumulated over one trillion tokens in usage.[10]

In November 2025, Volcano Engine launched Doubao-Seed-Code, a coding model integrated with the Trae app that scored 78.8% on SWE-Bench Verified and whose coding plan was priced about 63% below the market average.[54]

By late December 2025, Doubao had surpassed 100 million daily active users, according to people familiar with ByteDance's internal data, and QuestMobile counted 155 million weekly active users.[12][16] ByteDance noted that Doubao's user acquisition and marketing spend was the lowest among all ByteDance products that had ever reached the 100 million DAU milestone, attributing much of the growth to organic distribution through its ecosystem of apps including Douyin (the Chinese version of TikTok) and Toutiao.[12]

### Doubao 2.0 and the Agent Era (2026)

On February 14, 2026, ByteDance released Doubao 2.0, powered by the new Seed 2.0 foundation model family.[14] The release was positioned as an "agent-era" upgrade, shifting the product from question-and-answer interactions toward executing complex, multi-step real-world tasks autonomously.[15]

The Seed 2.0 family includes four model variants: Seed 2.0 Pro for frontier reasoning and complex agents, Seed 2.0 Lite for general production workloads, Seed 2.0 Mini for high-throughput batch processing, and Seed 2.0 Code for software development (optimized for the [TRAE](https://aiwiki.ai/wiki/trae) IDE).[19] The flagship Pro variant uses "System 2" thinking processes with advanced reasoning chains. ByteDance internally benchmarks Seed 2.0 Pro at parity with [GPT-5.2](https://aiwiki.ai/wiki/gpt-5) and [Gemini 3 Pro](https://aiwiki.ai/wiki/gemini) on math, coding, and logical reasoning.[14] Reported benchmark scores include 98.3 on [AIME 2025](https://aiwiki.ai/wiki/aime_2025), a 3020 Codeforces rating, 88.9 on [GPQA Diamond](https://aiwiki.ai/wiki/gpqa_diamond), 87.8 on LiveCodeBench v6, 76.5 on [SWE-Bench Verified](https://aiwiki.ai/wiki/swe-bench_verified), and 89.5 on VideoMME for hour-long video processing.[19] The Pro variant also reportedly earned gold medal scores on the ICPC, IMO (35 of 42 points), and CMO programming and mathematics competitions.[19] As of February 16, 2026, Seed 2.0 ranked sixth on the LMSYS Chatbot Arena Text leaderboard and third on the Vision Arena leaderboard.[19]

Pricing for Seed 2.0 Pro was set at approximately $0.47 per million input tokens and $2.37 per million output tokens, roughly 3.7 times cheaper on input and 5.9 times cheaper on output than [GPT-5.2](https://aiwiki.ai/wiki/gpt-5), and roughly 10 times cheaper than [Claude Opus 4.5](https://aiwiki.ai/wiki/claude_opus_4_5) (priced at $5.00 input / $25.00 output per million tokens).[20]

Two days later, on February 16, 2026 (Lunar New Year's Eve), ByteDance's Volcano Engine served as the exclusive AI cloud partner of the CCTV Spring Festival Gala, China's most-watched television broadcast, and the Doubao app was promoted heavily on screen.[18] During the live event, Doubao handled over 1.9 billion AI-related queries, and at peak demand it processed 63.3 billion tokens in a single minute after a host invited viewers to join an interactive segment.[18] Doubao users created more than 50 million themed avatars and sent over 100 million Lunar New Year greetings on the night.[18] The partnership also included three rounds of lottery during the broadcast and over 100,000 technology product giveaways spanning robots, drones, 3D printers, and usage rights for Audi and Mercedes-Benz electric vehicles, with individual cash prizes reaching as high as 8,888 yuan.[18] Doubao's gala campaign followed [Tencent](https://aiwiki.ai/wiki/tencent_ai) Yuanbao's 1 billion yuan coupon giveaway and Alibaba [Qwen](https://aiwiki.ai/wiki/qwen)'s 3 billion yuan promotion, in a holiday battle in which China's largest tech companies spent billions to acquire users.[17] Doubao's DAU surged to over 100 million on that day, roughly four times its early-February levels.[17]

Rankings have differed by tracker and methodology. In March 2025, Alibaba's revamped Quark assistant briefly overtook Doubao on Aicpb.com's ranking, with nearly 150 million monthly active users against almost 100 million for Doubao and around 77 million for DeepSeek.[21] By the first quarter of 2026, QuestMobile ranked Doubao first by a wide margin, with 345 million monthly active users, up 44.2% from the previous quarter and more than 180 million ahead of second-placed Qwen.[44]

Later estimates from QuestMobile, a Chinese mobile-internet data firm, put Doubao's monthly active users at 345 million in the first quarter of 2026 and 382 million in June 2026, and its weekly active users at 318 million in the week of June 22-28, 2026.[44][45][46] On QuestMobile's January-February 2026 daily-user chart, Doubao peaked at 144.55 million in mid-February, around the Spring Festival Gala.[46] Among the three apps QuestMobile tracks as typical AI advertisers (Doubao, Alibaba's Qwen and Tencent's Yuanbao), it estimated that Doubao accounted for 12.0% of their combined media (hard-ad) spending and 18.2% of their influencer (soft-ad) spending in June 2026, against 77.7% and 79.4% for Qwen.[46] In August 2026, 77.2% of China's AI-native app users opened Doubao during the month (54.9% a year earlier), and 12.4% of Doubao's monthly active users made a payment inside the app.[46] These are third-party modeled estimates; the [Qwen app](https://aiwiki.ai/wiki/tongyi_qianwen) article compares them across the leading apps.

## Underlying Technology

### Seed Model Family

Doubao is powered by ByteDance's Seed model family, developed by the company's Seed research team. The model lineage has evolved through several generations:

| Model | Release Date | Key Characteristics |
|---|---|---|
| Skylark (Yunque) | Mid-2023 | Original LLM, later renamed; powered the first version of Doubao |
| Doubao General Model Pro | May 2024 | First enterprise-grade model release on Volcengine |
| Doubao-1.5-Pro | January 2025 | Sparse MoE architecture, 20B active parameters, deep thinking mode |
| Seed-Thinking-v1.5 | April 2025 | 20B active / 200B total MoE reasoning model trained with VAPO and DAPO RL algorithms |
| Seed1.5-VL | May 2025 | Vision-language model, 532M [vision encoder](https://aiwiki.ai/wiki/vision_encoder) + 20B MoE LLM, SOTA on 38/60 benchmarks |
| Doubao 1.6-Vision | October 2025 | [Multimodal model](https://aiwiki.ai/wiki/multimodal_model), first in the family with tool-calling for visual reasoning, 50% cost reduction |
| Doubao-Seed-Code | November 2025 | Coding-specialized model, 78.8% on SWE-Bench Verified |
| Seed 1.8 | December 2025 | Generalized agentic model with GUI, search, and code agent capabilities |
| Seed 2.0 (Pro/Lite/Mini/Code) | February 2026 | Agent-era [foundation models](https://aiwiki.ai/wiki/foundation_models), long-chain reasoning, [GPT-5.2](https://aiwiki.ai/wiki/gpt-5.2) parity claimed |
| Seed 2.1 (Pro/Turbo) | June 2026 | Agent-capable productivity models; power Doubao's paid professional plans[48][56] |

The Seed 1.8 model introduced native foundational vision capabilities that allow it to directly interact with graphical interfaces across desktop, web, and mobile environments.[11] It supports three thinking modes and integrates a video tool (VideoCut) for long-video reasoning, scoring 87.8 on VideoMME.[11]

### Architecture

The Doubao model family uses a sparse Mixture of Experts (MoE) architecture, where only a subset of parameters are activated for each input. For example, Doubao-1.5-Pro activates 20 billion parameters per inference pass while maintaining performance comparable to a 140-billion-parameter dense model.[4] This approach allows ByteDance to offer high-performance AI at significantly reduced computational cost, which directly enables its aggressive pricing strategy.

The Seed 2.0 generation employs what Creati.ai describes as "System 2" thinking, referring to multi-step deliberative reasoning processes (as opposed to fast, pattern-matching "System 1" responses).[15] This architecture is designed for [agentic workflow](https://aiwiki.ai/wiki/agentic_workflow)s where the model must decompose complex objectives into executable sub-tasks and orchestrate multiple tools and API calls autonomously.[15]

### How does ByteDance train the Seed models?

A core differentiator of the Seed family is ByteDance's investment in reinforcement learning infrastructure. The Seed-Thinking-v1.5 paper details two in-house RL algorithms used during post-training: VAPO (Value-based Augmented Proximal Policy Optimization), an actor-critic method that stabilizes value estimates over long reasoning trajectories, and [DAPO (Decoupled Clip and Dynamic Sampling Policy Optimization)](https://aiwiki.ai/wiki/dapo), a policy-gradient method that addresses reward sparsity in chain-of-thought training.[5] The team also introduced new internal benchmarks and a blended reward system mixing verifiable outputs (for math and code) with non-verifiable outputs (for general tasks), reportedly contributing to the model's strong showing across both reasoning and conversational evaluations.[5] ByteDance has reported speeding up its reinforcement learning cycle roughly threefold through a HybridFlow framework and a streaming deployment system.[5]

## Features and Capabilities

Doubao provides a broad set of AI-powered features across its consumer app and enterprise API offerings.

### Feature Overview

| Feature | Description | Underlying Model |
|---|---|---|
| Text Chat | Conversational AI assistant for Q&A, brainstorming, writing, and general knowledge | Seed 2.0 Pro / Lite / Mini; Seed 2.1 from June 2026 |
| Deep Thinking | Extended reasoning mode for complex math, logic, and analysis problems | Seed 2.0 Pro (System 2 reasoning) |
| Image Generation | Text-to-image creation with support for Chinese aesthetics and 4K resolution | [Seedream](https://aiwiki.ai/wiki/seedream) 4.5 (up to 5.0 Lite) |
| Video Generation | Text/image/video-to-video creation with audio sync | [Seedance](https://aiwiki.ai/wiki/seedance) 2.0; Seedance 2.5 from July 2026 |
| Visual Understanding | Image and video analysis, chart interpretation, OCR, and visual Q&A | Seed1.5-VL / Doubao 1.6-Vision |
| Document Analysis | Upload and analyze PDFs, reports, and long-form documents; extract key points | Seed 2.0 with long-context support |
| Voice Interaction | Speech-to-text and [text-to-speech](https://aiwiki.ai/wiki/text_to_speech_ai) with emotional expression and dialect support | [Volcano Engine](https://aiwiki.ai/wiki/volcano_engine) speech models |
| Real-Time Voice Agent | Speech-to-speech assistant supporting interruption, multi-character performance, and 20+ Chinese dialects | End-to-end real-time voice model |
| Code Generation | Programming assistance, debugging, and code review | Seed 2.0 Code / Doubao-Seed-Code |
| AI Agent Tasks | Multi-step autonomous task execution (travel planning, research, GUI automation) | Seed 1.8 / Seed 2.0 Pro |
| OS-Level Agent | System-service agent on the Doubao Phone able to drive UI across third-party apps via voice | Seed 1.8 / Seed 2.0 Pro |
| Web Search | Real-time [information retrieval](https://aiwiki.ai/wiki/information_retrieval) integrated into chat responses | Integrated search capabilities |
| Translation | Support for 28+ languages with quality comparable to [GPT-4o](https://aiwiki.ai/wiki/gpt_4o) | Doubao translation model |

### Image Generation (Seedream)

Doubao's image generation is powered by the Seedream model family. Seedream 4.5, which Volcano Engine promoted at its December 2025 FORCE conference, generates images at up to 4K resolution and added multi-image combination and editing features.[37][50] The model has been specifically tuned for "Chinese aesthetics," producing artwork that reflects traditional Chinese artistic styles.

In February 2026, ByteDance released **[Seedream 5.0](https://aiwiki.ai/wiki/seedream_5)** (along with the lighter Seedream 5.0 Lite).[37] The 5.0 generation introduces chain-of-thought visual reasoning, which has the model plan composition, spatial layout, and physical accuracy before any pixels are rendered.[37] It also adds real-time web search so the model can pull live information into a generation request.[37] Seedream 5.0 Lite is positioned as a "unified multimodal image generation model" that combines deep thinking, online search, and generation in a single pipeline.[38]

### Video Generation (Seedance)

The Seedance model family handles video generation within Doubao. Seedance 1.0 was launched at the Volcengine FORCE conference on June 11, 2025, and ranked first on both the text-to-video and image-to-video leaderboards maintained by [Artificial Analysis](https://aiwiki.ai/wiki/artificial_analysis) as of June 10, 2025, generating a 5-second 1080p clip in about 41.4 seconds and beating [Veo 3](https://aiwiki.ai/wiki/veo_3) and Kling 2.0 by over 100 points on the image-to-video task.[43] [Seedance 2.0](https://aiwiki.ai/wiki/seedance_2), released in February 2026, accepts four input modalities in one request: text, up to 9 images, up to 3 video clips, and up to 3 audio clips, and generates video with synchronized audio. Its successor, Seedance 2.5, released on July 31, 2026, extends single-pass generation to 30 seconds and is available in the Doubao Pro app.[57]

Volcano Engine disclosed Seedance 2.0 pricing on March 4, 2026: pure [text-to-video generation](https://aiwiki.ai/wiki/text_to_video) costs 46 yuan ($6.40) per million tokens, while video-to-video editing mode costs 28 yuan ($3.90) per million tokens.[36] A 15-second clip consumes about 308,880 tokens, which works out to roughly 1 yuan ($0.14) per second of generated video.[35] When the pricing was published, Seedance 2.0 did not yet support API calls and was limited to a free experience quota on Volcano Engine;[36] Volcano Engine opened the Seedance 2.0 API by early April 2026.[39]

### Voice and Speech

Doubao supports voice-based interaction through both its mobile and desktop apps. Users can tap a microphone icon to speak instead of typing, and the app responds with synthesized speech. In December 2024, Volcengine announced plans for an end-to-end real-time voice model with capabilities including multi-character performance and dialect conversion.[55]

The current generation uses a speech-to-speech architecture, allowing the assistant to respond faster than traditional speech-to-text-to-speech pipelines and letting the user interrupt the assistant mid-sentence.[30] DataGlobeHub lists support for about 20 regional Chinese dialects, a differentiator for users who speak Cantonese, Sichuanese, Shanghainese, or other non-Mandarin varieties.[40]

### Document and Visual Analysis

Users can upload files (PDFs, images, spreadsheets) for analysis through the app's paperclip interface. Doubao can summarize long articles, extract key points from reports and meeting notes, interpret charts and graphs, and analyze video content. The Seed1.5-VL model provides the visual understanding backbone, excelling at complex reasoning, optical character recognition, image interpretation, and open-vocabulary detection.[7]

### OS Agent and Doubao Phone

In December 2025, ByteDance partnered with smartphone manufacturer ZTE to launch the Nubia M153, marketed as an "engineering prototype" for testing operating-system-level [agentic AI](https://aiwiki.ai/wiki/agentic_ai).[26] The device was announced in China on December 1, 2025, priced at 3,499 yuan (about $494), with a 6.78-inch screen and a Qualcomm Snapdragon 8 Elite Gen 5 processor.[26] It embeds Doubao directly at the operating-system level, where it runs as a system service rather than a normal app.[26] The OS agent can read on-screen content, drive third-party app interfaces, and execute multi-step natural language requests entirely by voice.[26] A ByteDance demo video showed the assistant identifying a product from a social media post, comparing prices across several e-commerce platforms, and placing an order on the cheapest one, with the user still completing payment.[28]

The first production run of the Doubao Phone (about 30,000 units) sold out quickly, and resellers listed units for about 1,000 yuan above retail.[28] ByteDance said it had no plans to make its own smartphones and was in talks with multiple phone makers about the assistant, and a public update log showed the handset receiving software releases about every two weeks.[27][29][30]

## Privacy, Security, and Censorship

Doubao operates under China's Interim Measures for the Management of Generative AI Services (effective August 2023), which require providers to register algorithms with the Cyberspace Administration of China, uphold "core socialist values," prevent generated content that undermines state authority or incites subversion, and label synthetic media.[34] Independent testing has found Doubao behaves consistently with these rules: it deflects, sanitizes, or declines to engage with politically sensitive topics such as the events of June 4, 1989, criticism of senior Chinese leadership, and Taiwanese sovereignty.[32] Taiwan's National Security Bureau, in a November 2025 review of Chinese AI models, reported that Doubao avoided politically charged terminology and exhibited bias on Taiwan-related queries.[33]

The launch of the Doubao Phone in December 2025 brought a separate set of privacy concerns. Because the OS-level agent has system permissions that go beyond what a normal app receives, including the ability to open virtually any installed application without going through a public API, a senior security executive at a major app told Nikkei Asia that Doubao had been granted excessively broad permissions.[28] Soon after launch, users reported that WeChat, Taobao, Alipay and several banking apps blocked or flagged the assistant, with some accounts frozen or logins refused; the companies attributed this to routine security controls.[28][29] The same executive said Beijing would not allow sensitive financial information to be managed by a private tech giant such as ByteDance.[28] ByteDance then restricted the assistant in banking, payment and some gaming scenarios, and the Doubao team said the assistant does not store on-screen content in the cloud or use it for model training.[29][51]

## Platform Availability

Doubao is available across multiple platforms:

- **Mobile:** iOS and Android apps with cloud synchronization
- **Desktop:** Native applications for Windows (10 and 11, including x86 and ARM) and macOS (10.15 Catalina and above, with native support for [Intel](https://aiwiki.ai/wiki/intel) and Apple Silicon M1/M2/M3 chips)
- **Web:** Fully functional browser-based version accessible through Chrome, Edge, and Safari at doubao.com
- **Smartphone OS integration:** Available as a system-level agent on the Nubia M153 Doubao Phone, with additional Chinese OEMs expected to follow

All platforms support account-based cloud synchronization, allowing users to access their conversations and settings across devices.

## International Version (Cici / Dola)

The international version of Doubao was originally launched as **Cici** in August 2023, targeting users outside mainland China.[23] Cici closely resembled Doubao in interface design, supported text and voice interaction, included image generation and analysis, and allowed users to interact with intelligent agents created by other users.[23]

In late 2025, ByteDance rebranded Cici to **Dola**. Existing users could log into Dola with the same account, with all previous chats and settings preserved.[24] The app is developed and distributed globally by SPRING (SG) PTE. LTD., a Singapore-based subsidiary of ByteDance.[24]

Dola is available in select markets including the United Kingdom, Mexico, Brazil, Indonesia, the Philippines, Malaysia, Thailand, Spain, France, Poland, the United Arab Emirates, Saudi Arabia, Egypt, and Argentina, but as of early 2026 was unavailable in the United States, Canada, and Australia.[24][40] The app supports roughly 18 international languages.[24][40] Some features available in the Chinese Doubao app, such as native audio-synced video generation, are blocked in the international version.[24] ByteDance does, however, license both the Seedream and Seedance models to international platforms through the Volcengine API, allowing developers outside China to access these capabilities without downloading the consumer app. By the end of 2025, Dola reportedly surpassed 10 million daily active users worldwide.[24][40]

## Coze Bot Builder Platform

Alongside Doubao, ByteDance operates **[Coze](https://aiwiki.ai/wiki/coze)**, a low-code platform for building custom AI chatbots and agents. ByteDance launched the Chinese version of Coze, known domestically as Kouzi, in February 2024, and also operates an international version.[52]

The platform provides a visual builder with drag-and-drop workflow design, plugin integrations (including [Google Search](https://aiwiki.ai/wiki/google_search), [DALL-E](https://aiwiki.ai/wiki/dall-e) 3, and CapCut), short-term and long-term memory systems, reusable UI cards, and deployment options to ecosystems such as Doubao, [Feishu](https://aiwiki.ai/wiki/feishu) (ByteDance's enterprise collaboration tool), and [Telegram](https://aiwiki.ai/wiki/telegram).[25] The domestic version runs on ByteDance's in-house Doubao models (with Doubao 1.5 Pro as the primary model), while the international version integrates external models such as [GPT-4o](https://aiwiki.ai/wiki/gpt_4o) and DALL-E.

On April 18, 2025, ByteDance began beta testing **Coze Space**, an agentic AI platform that combines workflow logic, collaborative planning, and access to proprietary data.[25]

In 2025, ByteDance open-sourced the core platform as **Coze Studio**, enabling developers to self-host and customize the bot-building infrastructure.

## Volcengine Cloud Integration

Doubao's models are commercially distributed through **Volcengine** (Volcano Engine), ByteDance's [cloud computing](https://aiwiki.ai/wiki/cloud_computing) and enterprise services platform. Volcengine offers the full Doubao model family through a Model-as-a-Service (MaaS) API, allowing enterprise customers to integrate Doubao's text, vision, code, and agent capabilities into their own products.

### Enterprise Adoption

As of December 2025, daily token consumption of Doubao models exceeded 50 trillion tokens, more than ten times the level a year earlier and 417 times the level at launch in May 2024, when Volcano Engine reported about 120 billion tokens a day.[2][10][50] Daily use reached 120 trillion tokens in March 2026[39] and more than 180 trillion by June 2026, according to Volcano Engine.[48] Over 100 enterprise customers had each accumulated more than one trillion tokens in total usage by year-end, a figure that grew to roughly 140 customers by March 2026, spanning industries including e-commerce, finance, entertainment, manufacturing, and education.[39]

In December 2025, Volcengine launched the **Doubao Assistant API**, making the core agent capabilities of the Doubao app (dialogue, reasoning, and search) directly available for enterprise integration. Future releases are planned to include multimodal understanding, in-depth research, content creation, and video call capabilities through the API. In April 2026, Volcengine became an official sponsor of the open-source **[OpenClaw](https://aiwiki.ai/wiki/openclaw)** agent framework and said it would co-build a Chinese mirror of OpenClaw's ClawHub skill repository; it also promoted ArkClaw, its service for running OpenClaw with Doubao models, and opened the Seedance 2.0 API.[39]

### Pricing Strategy

ByteDance has consistently positioned Doubao as one of the most cost-effective AI model offerings globally:

| Model / Period | Pricing | Comparison |
|---|---|---|
| Doubao enterprise model (May 2024 launch) | 0.0008 yuan per 1,000 tokens | 99.3% below industry average of 12 yuan cents per 1,000 tokens |
| Doubao-1.5-Pro (January 2025) | Comparable to [GPT-4o](https://aiwiki.ai/wiki/gpt_4o) quality | Approximately 50x cheaper than GPT-4o |
| Doubao Visual Understanding (December 2024) | 0.003 yuan per 1,000 tokens | 85% below industry average |
| Doubao-Seed-Code (November 2025) | RMB 9.9 for the first month of the coding plan | About 63% below the market average |
| Seed 2.0 Pro (February 2026) | $0.47/M input, $2.37/M output tokens | 3.7x cheaper input, 5.9x cheaper output vs. [GPT-5.2](https://aiwiki.ai/wiki/gpt-5.2); ~10x cheaper than [Claude Opus 4.5](https://aiwiki.ai/wiki/claude_opus_4_5) |
| [Seedance 2.0](https://aiwiki.ai/wiki/seedance_2) video (March 2026) | 46 yuan/M tokens (gen), 28 yuan/M tokens (edit) | About $0.14 per second of generated video |

This pricing strategy relies on the computational efficiency of the MoE architecture (activating only a fraction of total parameters per query) and ByteDance's massive infrastructure investments.

## Monetization

Doubao's monetization model differs sharply from that of Western consumer AI products. Where OpenAI generates the bulk of its revenue from individual and enterprise ChatGPT subscriptions, Doubao was entirely free to use for consumers in China until June 2026, and ByteDance derives the majority of its AI revenue from Volcengine's Model-as-a-Service business. A March 2026 Ainvest analysis argued that Doubao's lead in users had not yet produced a clear, profitable consumer business model, and that education-related in-app purchases were the main source of consumer AI app revenue in China.[41]

On June 24, 2026, ByteDance introduced paid subscriptions for Doubao, which AFP described as the first paid subscription service from a major Chinese AI chatbot; ByteDance said the professional version would offer higher usage quotas, and AFP reported that paid users could organize local files, build websites and generate slides in the app.[47] According to Rui Caijing, the professional version runs on the Doubao 2.1 model series and has three plans at 68, 200 and 500 yuan a month, with annual billing of up to 5,088 yuan, while free users can also try an office-task mode on the Doubao 2.1 Turbo model.[48]

Secondary monetization streams include in-app purchases for educational content (a category that, across Chinese AI apps, dominates direct consumer spend), licensing of the Seedream and Seedance models to international platforms via Volcengine, and indirect revenue from Doubao traffic that is funneled to other ByteDance properties such as Douyin and Toutiao.[41]

## Infrastructure Investment

ByteDance has committed substantial capital to [AI infrastructure](https://aiwiki.ai/wiki/ai_infrastructure). Reports indicate the company planned approximately $20 billion in capital expenditure for 2025, with a significant portion directed toward AI infrastructure development. As of December 2025, ByteDance had preliminarily budgeted RMB 160 billion (approximately $23 billion) in capital expenditures for 2026, with more than half allocated to procuring advanced semiconductors for AI model development.[22] The company has faced challenges related to U.S. export controls on advanced [GPUs](https://aiwiki.ai/wiki/gpu_computing), and The Information reported in January 2025 that it planned to spend about $7 billion that year to rent [NVIDIA](https://aiwiki.ai/wiki/nvidia) GPUs in data centers outside mainland China as a workaround.[58]

## Competition in China's AI Market

Doubao operates in an intensely competitive Chinese AI chatbot market. Its primary competitors include:

| Product | Developer | Notable metrics (2026) |
|---|---|---|
| Doubao | [ByteDance](https://aiwiki.ai/wiki/bytedance_ai) | 100M+ DAU (Feb 2026); 345M MAU in Q1 2026 (QuestMobile)[44] |
| [DeepSeek](https://aiwiki.ai/wiki/deepseek) | DeepSeek (High-Flyer) | 127M MAU in Q1 2026 (QuestMobile)[44] |
| [Ernie Bot](https://aiwiki.ai/wiki/baidu_ernie) (Wenxin Yiyan) | [Baidu](https://aiwiki.ai/wiki/baidu_ai) | Became free in April 2025 |
| [Qwen app](https://aiwiki.ai/wiki/tongyi_qianwen) (formerly Tongyi Qianwen) | [Alibaba](https://aiwiki.ai/wiki/alibaba_cloud) | 166M MAU in Q1 2026 (QuestMobile)[44] |
| Quark AI | Alibaba | Nearly 150M MAU, top of Aicpb.com's ranking in March 2025[21]; 83M weekly active users in late June 2026 (QuestMobile)[46] |
| [Yuanbao](https://aiwiki.ai/wiki/tencent_yuanbao) | [Tencent](https://aiwiki.ai/wiki/tencent_ai) | 57.35M MAU in Q1 2026 (QuestMobile)[44]; ran 1B yuan red-packet promo for 2026 Spring Festival |
| [Kimi](https://aiwiki.ai/wiki/moonshot_ai) | Moonshot AI | 8.34M MAU in Q1 2026, a fourth straight quarterly decline (QuestMobile)[44]; [Kimi K2](https://aiwiki.ai/wiki/kimi_k2) model (mid-2025) |
| [Zhipu AI](https://aiwiki.ai/wiki/zhipu_ai) ([ChatGLM](https://aiwiki.ai/wiki/chatglm)) | Zhipu AI | Enterprise-focused, open-source GLM models |

According to QuestMobile, China's AI apps had 446 million users as of March 2026, and the combined (non-deduplicated) monthly active users of the top five AI apps rose from 355 million in the third quarter of 2025 to 722 million in the first quarter of 2026, driven largely by Spring Festival promotions.[44]

Doubao's competitive advantages stem from ByteDance's distribution network (leveraging Douyin and Toutiao for organic user acquisition), aggressive pricing that undercuts competitors by an order of magnitude, and the breadth of its multimodal capabilities (text, image, video, voice, and code in a single app). The integration of Seedream for image generation and Seedance for video creation gives Doubao a content-creation edge that aligns with ByteDance's core competency in short-form video.

### How does Doubao compare to Quark and DeepSeek?

The competitive rankings have been fluid. In March 2025, Alibaba's Quark briefly overtook Doubao on Aicpb.com's MAU ranking after Alibaba repositioned Quark from a cloud storage and search app into an "AI super assistant" powered by Qwen [reasoning models](https://aiwiki.ai/wiki/reasoning_models); Aicpb counted app users and excluded direct visits to chatbot websites, which shows how methodology can swing rankings.[21] Doubao regained the top spot from April 2025 and, on QuestMobile's figures, widened its lead: in the week of June 22-28, 2026, it had 318 million weekly active users against 83 million for Quark.[40][46] Doubao's edge over DeepSeek widened through 2025: QuestMobile found Doubao reached 157 million MAU in August 2025 while DeepSeek slipped to roughly 143 million, with nearly 40% of departing DeepSeek users migrating to Doubao.[8][49]

## Key Milestones

| Date | Milestone |
|---|---|
| June 2023 | [ByteDance](https://aiwiki.ai/wiki/bytedance) begins internal testing of AI [chatbot](https://aiwiki.ai/wiki/chatbot) (codenamed Yunque) |
| August 2023 | Doubao launches publicly; international version Cici released |
| May 2024 | Doubao model family launched on Volcengine at 99.3% below industry pricing; app passes 100M downloads |
| November 2024 | Becomes China's most popular AI [chatbot](https://aiwiki.ai/wiki/chatbot) (~60M MAU) |
| December 2024 | Visual understanding and 3D generation models released at FORCE conference |
| January 2025 | Doubao-1.5-Pro released with deep thinking mode; video generation opens to public |
| April 2025 | Seed-Thinking-v1.5 reasoning model technical report and Volcengine API launch |
| May 2025 | Seed1.5-VL vision-language model released (SOTA on 38/60 benchmarks) |
| June 2025 | Seedance 1.0 video model launched, ranks first on both [Artificial Analysis](https://aiwiki.ai/wiki/artificial_analysis) leaderboards |
| August 2025 | 157M MAU, surpasses [DeepSeek](https://aiwiki.ai/wiki/deepseek) as top native AI app in China |
| December 2025 | Seed 1.8 agentic model and Seedance 1.5 Pro released; 50T daily tokens; 100M DAU reached; Doubao Phone (Nubia M153) launched with ZTE |
| February 2026 | Seed 2.0 / Doubao 2.0 released; [Seedream 5.0](https://aiwiki.ai/wiki/seedream_5) released; [Seedance 2.0](https://aiwiki.ai/wiki/seedance_2) released; CCTV Spring Festival Gala partnership; 1.9B queries and 63.3B tokens/minute peak on Lunar New Year's Eve |
| March 2026 | [Seedance 2.0](https://aiwiki.ai/wiki/seedance_2) pricing disclosed |
| April 2026 | Volcengine sponsors the open-source [OpenClaw](https://aiwiki.ai/wiki/openclaw) agent framework and opens the Seedance 2.0 API; ~140 enterprise customers exceed 1T tokens of cumulative usage |
| June 2026 | Seed 2.1 released; paid Doubao professional subscriptions launched; QuestMobile puts June MAU at 382M |
| July 2026 | Seedance 2.5 released and made available in Doubao Pro |

## See Also

- [ByteDance](https://aiwiki.ai/wiki/bytedance_ai)
- [Large Language Models](https://aiwiki.ai/wiki/large_language_model)
- [DeepSeek](https://aiwiki.ai/wiki/deepseek)
- [ChatGPT](https://aiwiki.ai/wiki/chatgpt)
- [Mixture of Experts](https://aiwiki.ai/wiki/mixture_of_experts)
- [Multimodal AI](https://aiwiki.ai/wiki/multimodal_ai)
- [Volcengine](https://aiwiki.ai/wiki/volcano_engine)
- [Seedream](https://aiwiki.ai/wiki/seedream)
- [Seedance](https://aiwiki.ai/wiki/seedance)

## References

1. "TikTok Parent Company ByteDance Releases [Generative AI](https://aiwiki.ai/wiki/generative_ai) Chatbot Dou Bao in China." Voicebot.ai, August 18, 2023. https://voicebot.ai/2023/08/18/tiktok-parent-company-bytedance-releases-generative-ai-chatbot-dou-bao-in-china/
2. "ByteDance's Volcano Engine Starts Price War in China's Enterprise LLMs Sector." Yicai Global, May 2024. https://www.yicaiglobal.com/news/bytedances-volcano-engine-starts-price-war-in-chinas-enterprise-llms-sector
3. "ByteDance Enters AI Arena with Doubao, Offering Ultra-Low Cost and Versatile Applications." Caixin Global, May 16, 2024. https://www.caixinglobal.com/2024-05-16/bytedance-enters-ai-arena-with-doubao-offering-ultra-low-cost-and-versatile-applications-102196732.html
4. "[ByteDance AI](https://aiwiki.ai/wiki/bytedance_ai) Introduces Doubao-1.5-Pro Language Model with a Deep Thinking Mode." MarkTechPost, January 25, 2025. https://www.marktechpost.com/2025/01/25/bytedance-ai-introduces-doubao-1-5-pro-language-model-with-a-deep-thinking-mode-and-matches-gpt-4o-and-claude-3-5-sonnet-benchmarks-at-50x-cheaper/
5. "Seed-Thinking-v1.5: Advancing Superb Reasoning Models with Reinforcement Learning." arXiv 2504.13914, April 2025. https://arxiv.org/abs/2504.13914
6. "Now it's TikTok parent ByteDance's turn for a reasoning AI: enter Seed-Thinking-v1.5." VentureBeat, April 2025. https://venturebeat.com/ai/now-its-tiktok-parent-bytedances-turn-for-a-reasoning-ai-enter-seed-thinking-v1-5
7. "Seed1.5-VL Technical Report." arXiv, May 2025. https://arxiv.org/abs/2505.07062
8. "QuestMobile Data: Doubao Monthly Active Users Exceed DeepSeek, Ranking First in China's Native AI App." AIBase, August 2025. https://www.aibase.com/news/www.aibase.com/news/21321
9. "Volcengine's Doubao 1.6-Vision Cuts Costs by 50% with New Tool-Calling Multimodal Model." TechNode, October 2, 2025. https://technode.com/2025/10/02/volcengines-doubao-1-6-vision-cuts-costs-by-50-with-new-tool-calling-multimodal-model/
10. "Volcano Engine FORCE Conference Unleashes: Doubao Large Model 1.8 + Seedance 1.5 Pro Released." AIBase News, December 2025. https://news.aibase.com/news/23817
11. "Official Release of Seed1.8: A Generalized Agentic Model." [ByteDance Seed](https://aiwiki.ai/wiki/bytedance_seed) Team, December 2025. https://seed.bytedance.com/en/blog/official-release-of-seed1-8-a-generalized-agentic-model
12. "ByteDance's Doubao reaches 100M DAU with minimal marketing spend." TechNode, December 25, 2025. https://technode.com/2025/12/25/bytedances-doubao-reaches-100m-dau-with-minimal-marketing-spend/
13. "36Kr Exclusive: Doubao's DAU Surpasses 100 Million." 36Kr, 2026. https://eu.36kr.com/en/p/3609313072153862
14. "ByteDance Releases Doubao-Seed-2.0, Positions Pro Model Against GPT 5.2 and Gemini 3 Pro." TechNode, February 14, 2026. https://technode.com/2026/02/14/bytedance-releases-doubao-seed-2-0-positions-pro-model-against-gpt-5-2-and-gemini-3-pro/
15. "ByteDance Unveils Doubao 2.0 AI Model for Complex Task Execution and [Agent](https://aiwiki.ai/wiki/agent) Era." Creati.ai, February 15, 2026. https://creati.ai/ai-news/2026-02-15/bytedance-unveils-doubao-2-0-ai-model-agent-era/
16. "China's ByteDance releases Doubao 2.0 AI model for 'agent era'." The Jakarta Post, February 15, 2026. https://www.thejakartapost.com/business/2026/02/15/chinas-bytedance-releases-doubao-20-ai-model-for-agent-era.html
17. "ByteDance's AI chatbot Doubao wins China holiday battle with 100 million users." Asharq Al-Awsat, February 2026. https://english.aawsat.com/technology/5244628-bytedances-ai-chatbot-doubao-wins-china-holiday-battle-100-million-users
18. "ByteDance's Doubao Dominates Spring Festival Gala With 1.9 Billion AI Interactions." Caixin Global, February 18, 2026. https://www.caixinglobal.com/2026-02-18/bytedances-doubao-dominates-spring-festival-gala-with-19-billion-ai-interactions-102415609.html
19. "Doubao Seed 2.0 Review: Benchmarks + Pricing." Evolink.ai, 2026. https://evolink.ai/blog/doubao-seed-2-0-review-benchmarks-pricing
20. "Doubao Seed 2.0 Pro Review: ByteDance's $0.47 Frontier Model (2026)." TokenMix Blog, 2026. https://tokenmix.ai/blog/doubao-seed-2-0-pro-review-2026
21. "Alibaba's Quark surpasses ByteDance's Doubao, DeepSeek as China's top AI app." South China Morning Post, March 2026. https://www.scmp.com/tech/big-tech/article/3306270/alibabas-quark-surpasses-bytedances-doubao-deepseek-chinas-top-ai-app
22. "ByteDance plans to lift AI spending to $23 bln in 2026, FT reports." Investing.com, December 2025. https://www.investing.com/news/stock-market-news/bytedance-plans-to-lift-ai-spending-to-23-bln-in-2026-ft-reports-4420503
23. "ByteDance's AI Assistant Cici Tops App Charts in Multiple Countries Abroad." AIBase News, 2025. https://news.aibase.com/news/22124
24. "Dola: The Rise of Doubao AI in the Global Market." Oreate AI Blog, 2026. https://www.oreateai.com/blog/dola-the-rise-of-doubao-ai-in-the-global-market/acf99160853b6ba5f012e3c846aa56cb
25. "ByteDance launches Coze, its new AI agent platform, in beta." KrASIA, April 24, 2025. https://kr-asia.com/bytedance-launches-coze-its-new-ai-agent-platform-in-beta
26. "ByteDance and ZTE Unveil 'Agentic' AI Smartphone Prototype." Winbuzzer, December 1, 2025. https://winbuzzer.com/2025/12/01/bytedance-and-zte-launch-agentic-ai-smartphone-prototype-xcxwbn/
27. "ByteDance pushes frequent updates to Doubao AI as privacy concerns persist." South China Morning Post, December 2025. https://www.scmp.com/tech/big-tech/article/3336773/bytedance-pushes-frequent-updates-doubao-ai-privacy-concerns-persist
28. "Why ByteDance's AI phone is getting blocked by China's biggest apps." Business Standard, December 2025. https://www.business-standard.com/world-news/bytedance-ai-phone-doubao-nubia-zte-app-blocks-user-control-china-125120800459_1.html
29. "ByteDance tweaks Doubao assistant amid app curbs." China Daily, December 9, 2025. https://www.chinadaily.com.cn/a/202512/09/WS69377cc2a310d6866eb2d997.html
30. "ByteDance Launches Doubao Real-Time AI Voice Assistant for Phones." Scientific American, 2025. https://www.scientificamerican.com/article/bytedance-launches-doubao-real-time-ai-voice-assistant-for-phones/
31. "China's AI chatbots are advanced and versatile and begging for more users." NPR, March 30, 2026. https://www.npr.org/2026/03/30/nx-s1-5760939/china-chatbot-industry-doubao-qwen-yuanbao
32. "Sensitive Prompts and Cultural Contexts: A Comparative Study of AI Chatbots in China and the West." Media and Journalism Research Center, September 2025. https://journalismresearch.org/2025/09/sensitive-prompts-and-cultural-contexts-a-comparative-study-of-ai-chatbots-in-china-and-the-west/
33. "China's AI models pose risks, NSB says." Taipei Times, November 17, 2025. https://www.taipeitimes.com/News/front/archives/2025/11/17/2003847325
34. "Shape of China's AI regulations and prospects." Law.asia, 2025. https://law.asia/china-ai-regulations-legislation-compliance-future-prospects/
35. "ByteDance's Seedance 2.0 video model costs about $0.14 per second." TechNode, March 5, 2026. https://technode.com/2026/03/05/bytedances-seedance-2-0-video-model-costs-about-0-14-per-second/
36. "Seedance 2.0 API Pricing: 46 CNY/Million Tokens (~1 CNY/Second)." AICost Blog, 2026. https://aicost.org/blog/seedance-2-0-api-pricing-breakdown-2026
37. "Seedream 5: Complete Guide to ByteDance's Next-Gen AI Image Generator (2026)." Createvision.ai, 2026. https://createvision.ai/guides/seedream-5-complete-guide
38. "How to Use Seedream 5.0 Lite API." EvoLink.AI, 2026. https://evolink.ai/blog/how-to-use-seedream-5-0-lite-api-2026
39. "ByteDance's Volcengine Powers AI Growth with OpenClaw Partnership." Caixin Global, April 2, 2026. https://www.caixinglobal.com/2026-04-02/bytedances-volcengine-powers-ai-growth-with-openclaw-partnership-102430344.html
40. "Doubao Statistics and Insights 2026." DataGlobeHub, 2026. https://dataglobehub.com/doubao-statistics-and-insights/
41. "Ranking the Chinese AI Apps: Aifu, DeepSeek, Qwen, Doubao, and the Scale-to-Revenue Gap." Ainvest.com, 2026. https://www.ainvest.com/news/ranking-chinese-ai-apps-aifu-deepseek-qwen-doubao-scale-revenue-gap-2603/
42. "What is Doubao? Complete Guide to Seedream & Seedance (2026)." PhotoGrid Blog, 2026. https://www.photogrid.app/blog/what-is-doubao/
43. "Seedance 1.0: Exploring the Boundaries of Video Generation Models." arXiv 2506.09113, June 2025. https://arxiv.org/abs/2506.09113
44. "近四成中国网民手机装AI，豆包月活逼近3.5亿，元宝跌出前三...2026年一季度AI应用价值榜" (2026 Q1 AI application value ranking). National Business Daily with QuestMobile, April 21, 2026. https://www.nbd.com.cn/articles/2026-04-21/4350446.html
45. "QuestMobile：豆包6月月活3.82亿 断层领跑AI原生APP榜" (QuestMobile first-half 2026 AI-native app ranking). Guandian, August 5, 2026. https://www.guandian.cn/article/20260805/581181.html
46. "QuestMobile 2026年AI用户洞察与营销分析报告" (QuestMobile 2026 AI user insight and marketing analysis report). QuestMobile, via 36Kr, October 13, 2026. https://www.36kr.com/p/4023659434512512
47. AFP. "ByteDance rolls out paid subscriptions for China's largest AI chatbot." The Star, June 25, 2026. https://www.thestar.com.my/tech/tech-news/2026/06/25/bytedance-rolls-out-paid-subscriptions-for-china039s-largest-ai-chatbot
48. Zhao Panpan. "豆包专业版正式收费，一年最高5088元" (Doubao professional version starts charging, up to 5,088 yuan a year). Rui Caijing, June 25, 2026. https://m.rccaijing.com/news-7475702713851508588.html
49. "ByteDance chatbot Doubao still China's most popular AI app as rival DeepSeek loses users." South China Morning Post, via Yahoo Tech, September 17, 2025. https://tech.yahoo.com/ai/articles/bytedance-chatbot-doubao-still-chinas-093000643.html
50. "发布以来增长417倍！火山引擎：豆包日均使用量超过50万亿" (Up 417-fold since launch: Volcano Engine says Doubao daily usage exceeds 50 trillion). Kuai Technology, via ifeng.com, December 18, 2025. https://tech.ifeng.com/c/8pBJAWLV2Ci
51. "WeChat, Chinese Banking Apps Block ByteDance's New AI Mobile Agent." Yicai Global, December 4, 2025. https://www2.yicaiglobal.com/news/wechat-some-chinese-banking-apps-reportedly-restrict-devices-with-bytedances-new-ai-voice-assistant
52. "ByteDance launches GPTs-like chatbot developing platform." TechNode, February 2, 2024. https://technode.com/2024/02/02/bytedance-launches-gpts-like-chatbot-developing-platform/
53. "Volcano Engine Releases Doubao 1.5 Deep Thinking Model." AIBase News, April 17, 2025. https://news.aibase.com/news/17246
54. "ByteDance unit launches Doubao-Seed-Code." KrASIA (summarizing South China Morning Post), November 12, 2025. https://kr-asia.com/pulses/158276
55. "ByteDance Adds Image Recognition, 3D Generation to Doubao AI Assistant." Yicai Global, December 2024. https://www.yicaiglobal.com/news/bytedance-unveils-doubao-visual-understanding-3d-generation-models
56. "Seed2.1 Officially Released: Advancing AI Productivity." ByteDance Seed, June 23, 2026. https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity
57. "One-take Creation, Flexible Referencing: Introducing Seedance 2.5." ByteDance Seed, July 31, 2026. https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5
58. "TikTok parent ByteDance plans to spend $7B on cloud-based GPUs this year to fuel its AI ambitions." SiliconANGLE, January 1, 2025. https://siliconangle.com/2025/01/01/tiktok-parent-bytedance-plans-spend-7b-cloud-based-gpus-year-fuel-ai-ambitions/