Most AI comparison tables answer a question nobody asked: which model scores highest. The question that actually decides which assistant you use is duller and more practical — can you install it, can you sign up for it, does it speak your language, and where do your conversations end up. This table leads with those, and it is honest about the one comparison it cannot make.
Doubao vs the world
Six assistants, side by side, on the things that decide whether you can actually use one: who makes it, what the free tier gives you, how good its English is, and where your conversations are processed.
| Doubao ByteDance (China) | DeepSeek Hangzhou DeepSeek (China) | ChatGPT OpenAI (US) | Grok SpaceXAI (US) | Qwen Alibaba (China) | Kimi Moonshot AI (China) | |
|---|---|---|---|---|---|---|
| Current flagship model | Doubao-Seed-2.1 Pro (released 23 June 2026) | DeepSeek-V4-Pro-0813 | GPT-5.6 Sol (GA 9 July 2026) | Grok 4.6 (released 12 August 2026) | Qwen3.8-Max (announced August 2026) | Kimi K3 |
| Can you install it outside China? | No. App Store: mainland China, Hong Kong and Macau only | Yes — iOS, Android and chat.deepseek.com | Yes — web, iOS, Android, desktop | Yes — web, iOS, Android | Yes — chat.qwen.ai | Yes — kimi.com and apps |
| English interface | No. Chinese is the only listed interface language | Yes | Yes | Yes | Yes | Yes |
| Free tier | Yes — search, writing, image generation, voice and video calls stay free | Yes — consumer chat is free, API billed separately | Yes — Free plan at $0 | Yes, with usage limits | Yes — web search, image generation and document analysis included | Yes, with paid tiers above it |
| Consumer paid plan | Doubao Pro, from CNY 68/month (launched 24 June 2026, mainland China only) | None published | Go $8, Plus $20, Pro $200 per month | Paid tiers offered via SpaceXAI | None published | $19 / $39 / $99 / $199 per month |
| Sign-up without a Chinese phone number | Only via Apple ID or a Douyin account — no email or Google option | Yes — email or Google | Yes — email, Google, Microsoft, Apple | Yes | Yes | Yes |
| Image generation | Yes — Seedream, built into the app | Not in the consumer chat | Yes | Yes | Yes | Not a headline feature |
| Voice / live video | Yes — real-time voice and video calls, upgraded to SeedRealtime in August 2026 | No voice mode in the consumer app | Yes — voice and video | Yes — voice mode | Yes — omni models cover audio and video | Limited |
| User-built agents | Withdrawn. Custom agent creation stopped on 15 July 2026 under China's anthropomorphic-AI rules | No | Yes — custom GPTs | Yes | Yes | Yes |
| Where your data is processed | Inside the People's Republic of China; the policy says data is not transferred overseas | Inside the People's Republic of China, including for EEA/UK users | United States and OpenAI's global infrastructure | United States | Beijing, Hong Kong, Singapore, Tokyo, Frankfurt or Virginia, depending on the region you use | Inside the People's Republic of China |
| Conversations used for training | Yes by default; opt out in Settings → Privacy and Permissions | Per its published policy | Per its published policy and account settings | Per its published policy | Per its published policy | Per its published policy |
| API price, flagship (per 1M tokens) | CNY 6 in / CNY 30 out on Volcengine; $0.50 in / $3.00 out for Seed 2.0 Pro on BytePlus | $0.66 in / $1.98 out off-peak, doubling at peak | $4 in / $20 out (gpt-5.6-sol) | $2 in / $6 out under 200K tokens | Published on Alibaba Cloud Model Studio | $3 in / $15 out (kimi-k3) |
| Context window | 256K tokens (Doubao-Seed-2.1 Pro) | Published per model | 1.05M tokens | 500K tokens | Up to 1M tokens | 1,048,576 tokens |
| On the public arena.ai leaderboard? | No — no Doubao or Seed model appears, so it cannot be ranked head-to-head | Yes — deepseek-v4-pro-high, 50th (preliminary) | Yes — gpt-5.6-sol-xhigh, 16th | Yes — grok-4.6-high, 46th (preliminary) | Yes — qwen3.8-max, 19th | Yes — kimi-k3-max, 10th |
Data reviewed 23 August 2026. Availability and features change without notice — always confirm against the official app store listing for your own country.
The row that matters most
Scroll to Can you install it outside China? and the table stops being a close contest. Doubao is the only one of the six that you cannot get. ByteDance lists the genuine app in three App Store storefronts — mainland China, Hong Kong and Macau — while DeepSeek, ChatGPT, Grok, Qwen and Kimi are all a download away in most of the world.
That is not a small caveat attached to an otherwise even comparison. It is the comparison. Doubao is China’s largest AI assistant by a wide margin — QuestMobile put it at 382 million monthly active users in June 2026, roughly 2.3 times second-placed Qwen — and almost none of those users are reading this page. Everything else in the table is context for readers deciding whether Doubao is worth pursuing at all.
What the table cannot tell you
There is no honest quality ranking here, and that is deliberate. As of August 2026, no Doubao or Seed model appears anywhere on the public arena.ai leaderboard, where the other five all place. Kimi K3 Max sits 10th, GPT-5.6 Sol 16th, Qwen3.8-Max 19th, Grok 4.6 and DeepSeek V4-Pro in the forties and fifties on preliminary scores. Doubao is simply absent.
That leaves only ByteDance’s own benchmark claims — a reported 54.2 on HLE-text for Doubao-Seed-2.0 Pro, parity with Claude Opus 4.7 on Terminal Bench 2.1 for 2.1 Pro, 78.80% on SWE-bench Verified for Doubao-Seed-Code — and vendor-reported numbers are not evidence of the same kind. We report them with the vendor’s name attached and we do not put them in a comparison table as though they were neutral measurements.
The table also says nothing about how these assistants behave on contested subjects. That is not a gap we can close with a cell in a grid; our safety article covers the documented research on it, including the Taiwanese government assessment and the academic comparison that tested Doubao against ChatGPT and Claude on the same politically sensitive prompts.
Reading the data-processing row
Three of the six process consumer data inside the People’s Republic of China. Doubao’s privacy policy states that information collected during domestic operations is stored inside China and that ByteDance does not currently transmit it overseas. DeepSeek’s English-language policy says it collects, processes and stores personal data in China — including for users in the EEA, the UK and Switzerland. Kimi’s policy says the same for its users.
Qwen is the interesting exception among the Chinese four: Alibaba Cloud’s Model Studio serves qwen3.8-max from Beijing, Hong Kong, Singapore, Tokyo, Frankfurt and US Virginia, so an international customer can have inference run outside mainland China. Whether that matters to you depends entirely on your threat model and your employer’s rules — it is a fact worth knowing, not a verdict.
One more row deserves attention because it moves the other way. Doubao’s privacy policy confirms that what you type, and what it replies, may be used to train the model — with an opt-out at Settings → Privacy and Permissions rather than an opt-in. That is the common arrangement for Chinese assistants, and it is the opposite of, for example, Anthropic’s consumer default, where training use requires you to switch it on.
The agents row is a story in itself
Doubao is the only assistant in this table whose user-built agents went backwards. On 15 July 2026 — the day China’s Interim Measures for the Administration of Anthropomorphic AI Interactive Services took effect — Doubao stopped letting users create custom agents in the app, and Alibaba’s Qwen posted an equivalent notice the same day. ByteDance pointed users to its separate Maobox app instead.
It is a good illustration of why a feature comparison between Chinese and Western assistants ages differently. Regulation is a product input in one market in a way it is not in the other. Our article on Doubao’s agents covers what remains.
How to use the toggles
Six columns is more than most screens want. Use the checkboxes above the table to hide assistants you are not considering — the Doubao column stays fixed, since it is the subject. The table is rendered as ordinary HTML, so it works with JavaScript disabled and reads correctly in a screen reader; the toggles are an enhancement on top, and nothing you do with them is stored or sent anywhere.
Every cell is traced to a primary source — official model documentation, published price lists, published privacy policies, or a named research firm with a date attached. Prices and model names in this category change monthly, so the date under the table is not decoration. For the head-to-head detail behind individual columns, see Doubao vs ChatGPT, Doubao vs DeepSeek and Doubao vs Grok, or start with what Doubao actually is.