Overview
Wenxin Yiyan, also known as 文心一言, is Baidu’s consumer AI assistant built on the ERNIE family of large language models. Launched in 2023, it serves Chinese-speaking users through a web and mobile app for chatting, writing, translation, coding, and document summarization, and it is tightly integrated with Baidu’s search index for grounded answers. For developers, the same models are available through Baidu AI Studio’s Qianfan platform as standard chat-completion APIs, so the assistant and the API share one model backbone. In our evaluation it is a strong default for Mandarin-language tasks and enterprise users already inside the Baidu ecosystem, with competitive reasoning on ERNIE 4.5 and 5.x. The main caveats are that its open-weight and global presence trail DeepSeek and Qwen, and pricing on the Qianfan API has risen in 2026, so heavy production use needs careful cost tracking.
Key Features
- Mandarin-first chat, writing, translation, and coding in one assistant
- ERNIE 4.5 / 5.x models with strong Chinese reasoning and long context
- Baidu search grounding for cited, up-to-date answers
- Qianfan API gives OpenAI-style access for developers
- Free app tier plus per-token API billing on Qianfan
Pricing
| Plan | Price | For |
|---|---|---|
| App (文心一言) | $0 | Free chat, writing, and search for consumers |
| ERNIE-Speed | Free | Permanent free API tier (QPS-limited) |
| ERNIE 4.5 Turbo | ~¥0.8/M in, ¥3.2/M out | Cheap production API, 128K context |
| ERNIE 5.1 | ~¥4/M in, ¥18/M out | Flagship reasoning API, per-token billing |
Comparison
vs. DeepSeek: DeepSeek open-weights its models and leads on global developer mindshare and price, while Wenxin Yiyan leans on Baidu’s search grounding and a polished consumer app. For English-heavy or open-weight needs, DeepSeek wins; for Mandarin search-backed answers, Wenxin holds up well. Pick DeepSeek to self-host; pick Wenxin for turnkey Chinese answers.
vs. Qwen: Qwen (Alibaba) ships a broader open model family and stronger coding, whereas Wenxin Yiyan’s edge is Baidu ecosystem integration. Both serve the Chinese market. Choose Qwen for open models and code; choose Wenxin for search-grounded consumer chat.
vs. Kimi: Kimi is built around very long context (256K) and document analysis, while Wenxin Yiyan is the broader generalist. Use Kimi for long PDFs and research; use Wenxin for everyday Chinese assistant tasks.